gpml-matlab-v3.1-2010-09-27/000755 000765 000024 00000000000 11527514620 015063 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/.octaverc000644 000765 000024 00000000010 11422240325 016651 0ustar00carlstaff000000 000000 startup gpml-matlab-v3.1-2010-09-27/gp.m000644 000765 000024 00000017036 11527514617 015664 0ustar00carlstaff000000 000000 function [varargout] = gp(hyp, inf, mean, cov, lik, x, y, xs, ys) % Gaussian Process inference and prediction. The gp function provides a % flexible framework for Bayesian inference and prediction with Gaussian % processes for scalar targets, i.e. both regression and binary % classification. The prior is Gaussian process, defined through specification % of its mean and covariance function. The likelihood function is also % specified. Both the prior and the likelihood may have hyperparameters % associated with them. % % Two modes are possible: training or prediction: if no test cases are % supplied, then the negative log marginal likelihood and its partial % derivatives w.r.t. the hyperparameters is computed; this mode is used to fit % the hyperparameters. If test cases are given, then the test set predictive % probabilities are returned. Usage: % % training: [nlZ dnlZ ] = gp(hyp, inf, mean, cov, lik, x, y); % prediction: [ymu ys2 fmu fs2 ] = gp(hyp, inf, mean, cov, lik, x, y, xs); % or: [ymu ys2 fmu fs2 lp] = gp(hyp, inf, mean, cov, lik, x, y, xs, ys); % % where: % % hyp column vector of hyperparameters % inf function specifying the inference method % cov prior covariance function (see below) % mean prior mean function % lik likelihood function % x n by D matrix of training inputs % y column vector of length n of training targets % xs ns by D matrix of test inputs % ys column vector of length nn of test targets % % nlZ returned value of the negative log marginal likelihood % dnlZ column vector of partial derivatives of the negative % log marginal likelihood w.r.t. each hyperparameter % ymu column vector (of length ns) of predictive output means % ys2 column vector (of length ns) of predictive output variances % fmu column vector (of length ns) of predictive latent means % fs2 column vector (of length ns) of predictive latent variances % lp column vector (of length ns) of log predictive probabilities % % post struct representation of the (approximate) posterior % 3rd output in training mode and 6th output in prediction mode % % See also covFunctions.m, infMethods.m, likFunctions.m, meanFunctions.m. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 if nargin<7 || nargin>9 disp('Usage: [nlZ dnlZ ] = gp(hyp, inf, mean, cov, lik, x, y);') disp(' or: [ymu ys2 fmu fs2 ] = gp(hyp, inf, mean, cov, lik, x, y, xs);') disp(' or: [ymu ys2 fmu fs2 lp] = gp(hyp, inf, mean, cov, lik, x, y, xs, ys);') return end if isempty(inf), inf = @infExact; else % set default inf if iscell(inf), inf = inf{1}; end % cell input is allowed if ischar(inf), inf = str2func(inf); end % convert into function handle end if isempty(mean), mean = {@meanZero}; end % set default mean if ischar(mean) || isa(mean, 'function_handle'), mean = {mean}; end % make cell if isempty(cov), error('Covariance function cannot be empty'); end % no default if ischar(cov) || isa(cov, 'function_handle'), cov = {cov}; end % make cell cov1 = cov{1}; if isa(cov1, 'function_handle'), cov1 = func2str(cov1); end if strcmp(cov1,'covFITC'); inf = @infFITC; end % only one possible inf alg if isempty(lik), lik = @likGauss; else % set default lik if iscell(lik), lik = lik{1}; end % cell input is allowed if ischar(lik), lik = str2func(lik); end % convert into function handle end D = size(x,2); if ~isfield(hyp,'mean'), hyp.mean = []; end % check the hyp specification if eval(feval(mean{:})) ~= numel(hyp.mean) error('Number of mean function hyperparameters disagree with mean function') end if ~isfield(hyp,'cov'), hyp.cov = []; end if eval(feval(cov{:})) ~= numel(hyp.cov) error('Number of cov function hyperparameters disagree with cov function') end if ~isfield(hyp,'lik'), hyp.lik = []; end if eval(feval(lik)) ~= numel(hyp.lik) error('Number of lik function hyperparameters disagree with lik function') end try % call the inference method % issue a warning if a classification likelihood is used in conjunction with % labels different from +1 and -1 if strcmp(func2str(lik),'likErf') || strcmp(func2str(lik),'likLogistic') uy = unique(y); if any( uy~=+1 & uy~=-1 ) warning('You attempt classification using labels different from {+1,-1}\n') end end if nargin>7 % compute marginal likelihood and its derivatives only if needed post = inf(hyp, mean, cov, lik, x, y); else if nargout==1 [post nlZ] = inf(hyp, mean, cov, lik, x, y); dnlZ = {}; else [post nlZ dnlZ] = inf(hyp, mean, cov, lik, x, y); end end catch msgstr = lasterr; if nargin > 7, error('Inference method failed [%s]', msgstr); else warning('Inference method failed [%s] .. attempting to continue',msgstr) dnlZ = struct('cov',0*hyp.cov, 'mean',0*hyp.mean, 'lik',0*hyp.lik); varargout = {NaN, dnlZ}; return % continue with a warning end end if nargin==7 % if no test cases are provided varargout = {nlZ, dnlZ, post}; % report -log marg lik, derivatives and post else alpha = post.alpha; L = post.L; sW = post.sW; if issparse(alpha) % handle things for sparse representations nz = alpha ~= 0; % determine nonzero indices if issparse(L), L = full(L(nz,nz)); end % convert L and sW if necessary if issparse(sW), sW = full(sW(nz)); end else nz = true(size(alpha)); end % non-sparse representation if numel(L)==0 % in case L is not provided, we compute it K = feval(cov{:}, hyp.cov, x(nz,:)); L = chol(eye(sum(nz))+sW*sW'.*K); end Ltril = all(all(tril(L,-1)==0)); % is L an upper triangular matrix? ns = size(xs,1); % number of data points nperbatch = 1000; % number of data points per mini batch nact = 0; % number of already processed test data points ymu = zeros(ns,1); ys2 = ymu; fmu = ymu; fs2 = ymu; lp = ymu; % allocate mem while nact use Cholesky parameters (alpha,sW,L) V = L'\(repmat(sW,1,length(id)).*Ks); fs2(id) = kss - sum(V.*V,1)'; % predictive variances else % L is not triangular => use alternative parametrisation fs2(id) = kss + sum(Ks.*(L*Ks),1)'; % predictive variances end fs2(id) = max(fs2(id),0); % remove numerical noise i.e. negative variances if nargin<9 [lp(id) ymu(id) ys2(id)] = lik(hyp.lik, [], fmu(id), fs2(id)); else [lp(id) ymu(id) ys2(id)] = lik(hyp.lik, ys(id), fmu(id), fs2(id)); end nact = id(end); % set counter to index of last processed data point end if nargin<9 varargout = {ymu, ys2, fmu, fs2, [], post}; % assign output arguments else varargout = {ymu, ys2, fmu, fs2, lp, post}; end end gpml-matlab-v3.1-2010-09-27/mean/000755 000765 000024 00000000000 11527514620 016003 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/util/000755 000765 000024 00000000000 11527514620 016040 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/likFunctions.m000644 000765 000024 00000007213 11527514620 017714 0ustar00carlstaff000000 000000 % likelihood functions are provided to be used by the gp.m function: % % likErf (Error function, classification, probit regression) % likLogistic (Logistic, classification, logit regression) % % likGauss (Gaussian, regression) % likLaplace (Laplacian or double exponential, regression) % likSech2 (Sech-square, regression) % likT (Student's t, regression) % % The likelihood functions have three possible modes, the mode being selected % as follows (where "lik" stands for any likelihood function in "lik/lik*.m".): % % 1) With one or no input arguments: [REPORT NUMBER OF HYPERPARAMETERS] % % s = lik OR s = lik(hyp) % % The likelihood function returns a string telling how many hyperparameters it % expects, using the convention that "D" is the dimension of the input space. % For example, calling "likLogistic" returns the string '0'. % % % 2) With three or four input arguments: [PREDICTION MODE] % % lp = lik(hyp, y, mu) OR [lp, ymu, ys2] = lik(hyp, y, mu, s2) % % This allows to evaluate the predictive distribution. Let p(y_*|f_*) be the % likelihood of a test point and N(f_*|mu,s2) an approximation to the posterior % marginal p(f_*|x_*,x,y) as returned by an inference method. The predictive % distribution p(y_*|x_*,x,y) is approximated by. % q(y_*) = \int N(f_*|mu,s2) p(y_*|f_*) df_* % % lp = log( q(y) ) for a particular value of y, if s2 is [] or 0, this % corresponds to log( p(y|mu) ) % ymu and ys2 the mean and variance of the predictive marginal q(y) % note that these two numbers do not depend on a particular % value of y % All vectors have the same size. % % % 3) With five or six input arguments, the fifth being a string [INFERENCE MODE] % % [varargout] = lik(hyp, y, mu, s2, inf) OR % [varargout] = lik(hyp, y, mu, s2, inf, i) % % There are three cases for inf, namely a) infLaplace, b) infEP and c) infVB. % The last input i, refers to derivatives w.r.t. the ith hyperparameter. % % a1) [sum(lp),dlp,d2lp,d3lp] = lik(hyp, y, f, [], 'infLaplace') % lp, dlp, d2lp and d3lp correspond to derivatives of the log likelihood % log(p(y|f)) w.r.t. to the latent location f. % lp = log( p(y|f) ) % dlp = d log( p(y|f) ) / df % d2lp = d^2 log( p(y|f) ) / df^2 % d3lp = d^3 log( p(y|f) ) / df^3 % % a2) [lp_dhyp,d2lp_dhyp] = lik(hyp, y, f, [], 'infLaplace', i) % returns derivatives w.r.t. to the ith hyperparameter % lp_dhyp = d log( p(y|f) ) / (df dhyp_i) % d2lp_dhyp = d^3 log( p(y|f) ) / (df^2 dhyp_i) % % % b1) [lZ,dlZ,d2lZ] = lik(hyp, y, mu, s2, 'infEP') % let Z = \int p(y|f) N(f|mu,s2) df then % lZ = log(Z) % dlZ = d log(Z) / dmu % d2lZ = d^2 log(Z) / dmu^2 % % b2) [dlZhyp] = lik(hyp, y, mu, s2, 'infEP', i) % returns derivatives w.r.t. to the ith hyperparameter % dlZhyp = d log(Z) / dhyp_i % % % c1) [h,b,dh,db,d2h,d2b] = lik(hyp, y, [], ga, 'infVB') % ga is the variance of a Gaussian lower bound to the likelihood p(y|f). % p(y|f) \ge exp( b*f - f.^2/(2*ga) - h(ga)/2 ) \propto N(f|b*ga,ga) % The function returns the linear part b and the "scaling function" h(ga) and % derivatives dh = d h/dga, db = d b/dga, d2h = d^2 h/dga and d2b = d^2 b/dga. % % c2) [dhhyp] = lik(hyp, y, [], ga, 'infVB', i) % dhhyp = dh / dhyp_i, the derivative w.r.t. the ith hyperparameter % % Cumulative likelihoods are designed for binary classification. Therefore, they % only look at the sign of the targets y; zero values are treated as +1. % % See the help for the individual likelihood for the computations specific to % each likelihood function. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 gpml-matlab-v3.1-2010-09-27/covFunctions.m000644 000765 000024 00000011406 11527514620 017723 0ustar00carlstaff000000 000000 % covariance functions to be use by Gaussian process functions. There are two % different kinds of covariance functions: simple and composite: % % simple covariance functions: % covConst - covariance for constant functions % covLIN - linear covariance function % covLINard - linear covariance function with ARD % covLINone - linear covariance function with bias % covMaterniso - Matern covariance function with nu=1/2, 3/2 or 5/2 % covNNone - neural network covariance function % covNoise - independent covariance function (i.e. white noise) % covPeriodic - smooth periodic covariance function (1d) with unit period % covPoly - polynomial covariance function % covPPiso - piecewise polynomial covariance function (compact support) % covRQard - rational quadratic covariance function with ARD % covRQiso - isotropic rational quadratic covariance function % covSEard - squared exponential covariance function with ARD % covSEiso - isotropic squared exponential covariance function % covSEisoU - as above but without latent scale % % composite (meta) covariance functions (see explanation at the bottom): % covScale - scaled version of a covariance function % covProd - products of covariance functions % covSum - sums of covariance functions % covADD - additive covariance function % covMask - mask some dimensions of the data % % special purpose (wrapper) covariance functions % covFITC - to be used in conjunction with infFITC for large scale % regression problems; any covariance can be wrapped by % covFITC such that the FITC approximation is applicable % % Naming convention: all covariance functions are named "cov/cov*.m". A trailing % "iso" means isotropic, "ard" means Automatic Relevance Determination, and % "one" means that the distance measure is parameterized by a single parameter. % % The covariance functions are written according to a special convention where % the exact behaviour depends on the number of input and output arguments % passed to the function. If you want to add new covariance functions, you % should follow this convention if you want them to work with the function gp. % There are four different ways of calling the covariance functions: % % 1) With no (or one) input argument(s): % % s = covNAME % % The covariance function returns a string s telling how many hyperparameters it % expects, using the convention that "D" is the dimension of the input space. % For example, calling "covRQard" returns the string '(D+2)'. % % 2) With two input arguments: % % K = covNAME(hyp, x) equivalent to K = covNAME(hyp, x, []) % % The function computes and returns the covariance matrix where hyp are % the hyperparameters and x is an n by D matrix of cases, where % D is the dimension of the input space. The returned covariance matrix is of % size n by n. % % 3) With three input arguments: % % Ks = covNAME(hyp, x, xs) % kss = covNAME(hyp, xs, 'diag') % % The function computes test set covariances; kss is a vector of self covariances % for the test cases in xs (of length ns) and Ks is an (n by ns) matrix of cross % covariances between training cases x and test cases xs. % % 4) With four input arguments: % % dKi = covNAME(hyp, x, [], i) % dKsi = covNAME(hyp, x, xs, i) % dkssi = covNAME(hyp, xs, 'diag', i) % % The function computes and returns the partial derivatives of the % covariance matrices with respect to hyp(i), i.e. with % respect to the hyperparameter number i. % % Covariance functions can be specified in two ways: either as a string % containing the name of the covariance function or using a cell array. For % example: % % cov = 'covRQard'; % cov = {'covRQard'}; % cov = {@covRQard}; % % are supported. Only the second and third form using the cell array can be used % for specifying composite covariance functions, made up of several % contributions. For example: % % cov = {'covScale', {'covRQiso'}}; % cov = {'covSum', {'covRQiso','covSEard','covNoise'}}; % cov = {'covProd',{'covRQiso','covSEard','covNoise'}}; % cov = {'covMask',{mask,'covSEiso'}} % q=1; cov = {'covPPiso',q}; % d=3; cov = {'covPoly',d}; % cov = {'covADD',{[1,2],'covSEiso'}}; % cov = {@covFITC, {@covSEiso}, u}; where u are the inducing inputs % % specifies a covariance function which is the sum of three contributions. To % find out how many hyperparameters this covariance function requires, we do: % % feval(cov{:}) % % which returns the string '3+(D+1)+1' (i.e. the 'covRQiso' contribution uses % 3 parameters, the 'covSEard' uses D+1 and 'covNoise' a single parameter). % % See also doc/usageCov.m. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 gpml-matlab-v3.1-2010-09-27/README000644 000765 000024 00000022560 11527514620 015750 0ustar00carlstaff000000 000000 GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox version 3.1 for GNU Octave 3.2.x and Matlab 7.x Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-27. 0) HOW TO READ ============== If you want to get started without further delay, then read section 1) below and jump right to the examples in doc/index.html. 1) ABOUT THESE PROGRAMS ======================= This collection of matlab programs implements and demonstrates some of the algorithms described in a) the book by Rasmussen and Williams: "Gaussian Processes for Machine Learning", the MIT Press 2006, in b) the article by Nickisch and Rasmussen: "Approximations for Binary Gaussian Process Classification", JMLR 2008 and in c) the article by Candela and Rasmussen: "A Unifying View of Sparse Approximate Gaussian Process Regression", JMLR 2005. There are 6 code subdirectories: cov, doc, inf, lik, mean and util. cov/ contains covariance functions cov*.m => see covFunctions.m doc/ contains an index.html file providing documentation. This information is also available from http://www.GaussianProcess.org/gpml/code. Usage of mean, cov, classification and regression is demonstrated in usage*.m. Further details can be found in the developer documentation manual.pdf. inf/ contains the inference algorithms inf*.m => see infMethods.m lik/ contains the likelihood functions lik*.m => see likFunctions.m mean/ contains the mean functions mean*.m => see meanFunctions.m util/ contains optimisation routines, backward compatibility programs and small auxiliary scripts Before running the demos or any other part of the code, you should execute startup.m to add all necessary directories to your path. In Octave, you can also use the .octaverc file. 2) ABOUT MEX FILES ================== Some of the programs make use of the MEX facility in matlab for more efficient implementation. However, if you don't know about how to compile MEX files, you do not need to worry about this -- the code runs anyway. If you do compile the MEX files, this is automatically detected, and the program will run more efficiently. If you do not have a compiler installed on your system, you might want to consult [2]. Two components of the toolbox can be accelerated by MEX. First, there is an auxiliary functions solve_chol.m having a MEX equivalent. This can be compiled by either executing util/make.m from the Matlab/Octave command line. In addition to that, we provide a Makefile for Matlab in util/ that you can run by simply typing make in your shell after having supplied your Matlab path and your operating system. We recommend the make.m script because it works both under Matlab and Octave. Second, in order to use the L-BFGS minimiser for hyperparameter optimisation as an equivalent to minimize.m, you have to compile Peter Carbonetto's "Matlab interface for L-BFGS-B" [1]. The challenge here is the Fortran 77 code. We provide a Makefile suitable for Linux 32/64 bit and Mac whenever you have 1) g77 or 2) gfortran properly set up. Under Ubuntu, you can achieve this by installing the packages fort77 and gfortran, respectively. Compilation is done by editing util/lbfgsb/Makefile. In any case, you need to provide $MATLAB_HOME which can be found by the commands 'locate matlab' or 'find / -name "matlab"'. You can choose between two compilation modes: a) using the mex utility by Matlab [default] provide $MEX, then type 'make mex' b) without mex utility by Matlab provide $MEX_SUFFIX and $MATLAB_LIB, then type 'make nomex' In Ubuntu 10.04 LTS, the libg2c library needed for both 1)+a) and 1)+b) is not included per default. If 'ls /usr/lib/libg2c.*' does not list anything this is the case on your machine. You then whant to install the packages gcc-3.4-base and libg2c0 e.g. from http://packages.ubuntu.com/hardy/gcc-3.4-base and http://packages.ubuntu.com/hardy/libg2c0. After installation, you have to create a symbolic link by 'cd /usr/lib' and 'ln -s libg2c.so.0 libg2c.so'. [1] http://www.cs.ubc.ca/~pcarbo/lbfgsb-for-matlab.html [2] http://www.mathworks.com/support/compilers/R2010a 3) CURRENT VERSION ================== The current version of the programs is 3.0. Previous versions of the code are available at http://www.gaussianprocess.org/gpml/code/oldcode.html. 4) DIFFERENCES TO PREVIOUS VERSIONS =================================== NEW in version 3.1, 2010-10-10 ------------------------------ - following a suggestion by Ed Snelson we now support FITC regression - cov/covFITC.m and inf/infFITC.m have been added with Ed Snelson's help - the covariance interface was slightly changed to make that possible NEW in version 3.0, 2010-07-23 ------------------------------ A major code reorganisation effort did take place in the current release. First, classification and regression are now done by a single file gp.m which is completely generic in the likelihood. The previous regression program gpr.m corresponds to gp.m with Gaussian likelihood. Several other likelihoods for robust regresssion were added. Further, the code now supports mean functions with a similar specification mechanism as already used by the covariance functions. Previous implementations correspond to using meanZero.m. We merged the covariance functions covMatern3iso.m and covMatern5.iso into a single covMaterniso.m and added a covariance function for additive functions as well as the possibility to use only certain components of the data. Finally, we included covPPiso.m a piecewise polynomial covariance function with compact support. New likelihood functions in lik/: likGauss.m, likLaplace.m, likLogistic.m and likT.m New mean functions in mean/: meanConst.m, meanLinear.m, meanPow.m, meanProd.m, meanSum.m and meanZero.m New covariance functions in cov/: covADD.m, covMask.m, covPPiso.m The gprSRPP.m function which previously provided "Subset of Regressors" and the "Projected Process" approximation has now been removed. NEW in version 2.1, 2007-07-25 ------------------------------ covConst.m: fixed a bug which caused an error in the derivative of the log marginal likelihood for certain combinations of covariance functions and approximation methods. (Thanks to Antonio Eleuteri for reporting the problem) gauher.m: added the function "gauher.m" which was mistakenly missing from the previous release. This caused an error for certain combinations of approximation method and likelihood function. logistic.m: modified the approximation of moments calculation to use a mixture of cumulative Gaussian, rather than Gauss-Hermite quadrature, as the former turns out to be more accurate. NEW in version 2.0, 2007-06-25 ------------------------------ Some code restructuring has taken place for the classification code to make it more modular, to facilitate addition of new likelihood functions and approximations methods. Now, all classification is done using the binaryGP function, which (among other things) takes an approximation method and a likelihood function as an arguments. Thus, binaryGP replaces both binaryEPGP and binaryLapaceGP, although wrapper functions are still provided for backward compatibility. This gives added flexibility: now EP can also be used wth the logistic likelihood function (implemented using Gauss-Hermite quadrature). approxEP.m: New file, containing the Expectation Propagation approximation method, which was previously contained in binaryEPGP.m approxLA.m: New file, containing Laplaces approximation method, which was previously contained in binaryLaplace.m approximations.m: New file, help for the approximation methods. binaryEPGP.m: This file has been replaced by a wrapper (for backward compatibility) which calls the more general binaryGP function. binaryGP.m: New general function to do binary classification. binaryLaplaceGP.m: This file has been replaced by a wrapper (for backward compatibility) which calls the more general binaryGP function. covMatern3iso.m, covMatern5iso.m, covNNone.m, covRQard.m, covRQiso.m, cosSEard, covSEiso: now check more carefully, that persistent variables have the correct sizes, and some variable names have been modified. cumGauss.m: New file, containing code for the cumulative Gaussian likelihood function likelihoods.m: New file, help for likelihood functions logistic.m: New file, logistic likelihood NEW in version 1.3, 2006-09-08 ------------------------------ covRQard.m: bugfix: replaced x with x' and z with z' in line 36 covRQiso.m: bugfix: replaced x with x' and z with z' in line 28 minimize.m: correction: replaced "error()" with "error('')", and made a few cosmetic changes binaryEPGP.m: added the line "lml = -n*log(2);" in line 77. This change should be largely inconsequential, but occationally may save things when the covariance matrix is exceptionally badly conditioned. NEW in version 1.2, 2006-05-10 ------------------------------ added the "erfint" function to "binaryLaplaceGP.m". The erfint function was missing by mistake, preventing the use of the "logistic" likelihood. NEW in version 1.1, 2006-04-12 ------------------------------ added files: "covProd.m" and "covPeriodic.m" changes: "covSEiso.m" was changed slightly to avoid the use of persistent variables NEW in version 1.0, 2006-03-29 ------------------------------ initial version shipped with the book gpml-matlab-v3.1-2010-09-27/inf/000755 000765 000024 00000000000 11527514620 015637 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/infMethods.m000644 000765 000024 00000003733 11527514617 017355 0ustar00carlstaff000000 000000 % Inference methods: Compute the (approximate) posterior for a Gaussian process. % Methods currently implemented include: % % infExact Exact inference (only possible with Gaussian likelihood) % infFITC Large scale regression with approximate covariance matrix % infLaplace Laplace's Approximation % infEP Expectation Propagation % infVB Variational Bayes % % infLOO Leave-One-Out predictive probability and Least-Squares Approxim. % % The interface to the approximation methods is the following: % % function [post nlZ dnlZ] = inf..(hyp, cov, lik, x, y) % % where: % % hyp is a struct of hyperparameters % cov is the name of the covariance function (see covFunctions.m) % lik is the name of the likelihood function (see likFunctions.m) % x is a n by D matrix of training inputs % y is a (column) vector (of size n) of targets % % nlZ is the returned value of the negative log marginal likelihood % dnlZ is a (column) vector of partial derivatives of the negative % log marginal likelihood w.r.t. each hyperparameter % post struct representation of the (approximate) posterior containing % alpha is a (sparse or full column vector) containing inv(K)*m, where K % is the prior covariance matrix and m the approx posterior mean % sW is a (sparse or full column) vector containing diagonal of sqrt(W) % the approximate posterior covariance matrix is inv(inv(K)+W) % L is a (sparse or full) matrix, L = chol(sW*K*sW+eye(n)) % % Usually, the approximate posterior to be returned admits the form % N(m=K*alpha, V=inv(inv(K)+W)), where alpha is a vector and W is diagonal; % if not, then L contains instead -inv(K+inv(W)), and sW is unused. % % For more information on the individual approximation methods and their % implementations, see the separate inf??.m files. See also gp.m % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 gpml-matlab-v3.1-2010-09-27/meanFunctions.m000644 000765 000024 00000003112 11527514620 020047 0ustar00carlstaff000000 000000 % mean functions to be use by Gaussian process functions. There are two % different kinds of mean functions: simple and composite: % % simple mean functions: % % meanZero - zero mean function % meanOne - one mean function % meanConst - constant mean function % meanLinear - linear mean function % % composite covariance functions (see explanation at the bottom): % % meanScale - scaled version of a mean function % meanPow - power of a mean function % meanProd - products of mean functions % meanSum - sums of mean functions % meanMask - mask some dimensions of the data % % Naming convention: all mean functions are named "mean/mean*.m". % % % 1) With no or only a single input argument: % % s = meanNAME or s = meanNAME(hyp) % % The mean function returns a string s telling how many hyperparameters hyp it % expects, using the convention that "D" is the dimension of the input space. % For example, calling "meanLinear" returns the string 'D'. % % 2) With two input arguments: % % m = meanNAME(hyp, x) % % The function computes and returns the mean vector where hyp are the % hyperparameters and x is an n by D matrix of cases, where D is the dimension % of the input space. The returned mean vector is of size n by 1. % % 3) With three input arguments: % % dm = meanNAME(hyp, x, i) % % The function computes and returns the n by 1 vector of partial derivatives % of the mean vector w.r.t. hyp(i) i.e. hyperparameter number i. % % See also doc/usageMean.m. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 gpml-matlab-v3.1-2010-09-27/lik/000755 000765 000024 00000000000 11527514620 015642 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/doc/000755 000765 000024 00000000000 11527514620 015630 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/Copyright000644 000765 000024 00000003375 11450142672 016765 0ustar00carlstaff000000 000000 GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox version 3.1 for GNU Octave 3.2.x and Matlab 7.x Copyright (c) 2005-2010 Carl Edward Rasmussen & Hannes Nickisch. All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. THIS SOFTWARE IS PROVIDED BY CARL EDWARD RASMUSSEN & HANNES NICKISCH ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL CARL EDWARD RASMUSSEN & HANNES NICKISCH OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. The views and conclusions contained in the software and documentation are those of the authors and should not be interpreted as representing official policies, either expressed or implied, of Carl Edward Rasmussen & Hannes Nickisch. The code and associated documentation is available from http://gaussianprocess.org/gpml/code. gpml-matlab-v3.1-2010-09-27/cov/000755 000765 000024 00000000000 11527514620 015652 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/startup.m000644 000765 000024 00000001410 11422240325 016727 0ustar00carlstaff000000 000000 % startup script to make Octave/Matlab aware of the GPML package % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-07-05. disp(['executing gpml startup script...']); OCT = exist('OCTAVE_VERSION') ~= 0; % check if we run Matlab or Octave me = mfilename; % what is my filename mydir = which(me); mydir = mydir(1:end-2-numel(me)); % where am I located if OCT && numel(mydir)==2 if strcmp(mydir,'./'), mydir = [pwd,mydir(2:end)]; end end % OCTAVE 3.0.x relative, MATLAB and newer have absolute path addpath(mydir(1:end-1)) addpath([mydir,'cov']) addpath([mydir,'doc']) addpath([mydir,'inf']) addpath([mydir,'lik']) addpath([mydir,'mean']) addpath([mydir,'util']) clear me mydir gpml-matlab-v3.1-2010-09-27/cov/covLIN.m000644 000765 000024 00000002036 11447142675 017173 0ustar00carlstaff000000 000000 function K = covLIN(hyp, x, z, i) % Linear covariance function. The covariance function is parameterized as: % % k(x^p,x^q) = x^p'*x^q % % The are no hyperparameters: % % hyp = [ ] % % Note that there is no bias or scale term; use covConst to add these. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '0'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode % compute inner products if dg % vector kxx K = sum(x.*x,2); else if xeqz % symmetric matrix Kxx K = x*x'; else % cross covariances Kxz K = x*z'; end end if nargin>3 % derivatives error('Unknown hyperparameter') endgpml-matlab-v3.1-2010-09-27/cov/covLINard.m000644 000765 000024 00000003011 11447142675 017654 0ustar00carlstaff000000 000000 function K = covLINard(hyp, x, z, i) % Linear covariance function with Automatic Relevance Determination (ARD). The % covariance function is parameterized as: % % k(x^p,x^q) = x^p'*inv(P)*x^q % % where the P matrix is diagonal with ARD parameters ell_1^2,...,ell_D^2, where % D is the dimension of the input space. The hyperparameters are: % % hyp = [ log(ell_1) % log(ell_2) % .. % log(ell_D) ] % % Note that there is no bias term; use covConst to add a bias. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = 'D'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode ell = exp(hyp); [n,D] = size(x); x = x*diag(1./ell); % precompute inner products if dg % vector kxx K = sum(x.*x,2); else if xeqz % symmetric matrix Kxx K = x*x'; else % cross covariances Kxz z = z*diag(1./ell); K = x*z'; end end if nargin>3 % derivatives if i<=D if dg K = -2*x(:,i).*x(:,i); else if xeqz K = -2*x(:,i)*x(:,i)'; else K = -2*x(:,i)*z(:,i)'; end end else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covPoly.m000644 000765 000024 00000002717 11447142675 017502 0ustar00carlstaff000000 000000 function K = covPoly(d, hyp, x, z, i) % Polynomial covariance function. The covariance function is parameterized as: % % k(x^p,x^q) = sf^2 * ( c + (x^p)'*(x^q) )^d % % The hyperparameters are: % % hyp = [ log(c) % log(sf) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<3, K = '2'; return; end % report number of parameters if nargin<4, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode c = exp(hyp(1)); % inhomogeneous offset sf2 = exp(2*hyp(2)); % signal variance if d~=max(1,fix(d)), error('only nonzero integers allowed for d'), end % degree % precompute inner products if dg % vector kxx K = sum(x.*x,2); else if xeqz % symmetric matrix Kxx K = x*x'; else % cross covariances Kxz K = x*z'; end end if nargin<5 % covariances K = sf2*( c + K ).^d; else % derivatives if i==1 K = c*d*sf2*( c + K ).^(d-1); elseif i==2 K = 2*sf2*( c + K ).^d; else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covLINone.m000644 000765 000024 00000002516 11447142675 017700 0ustar00carlstaff000000 000000 function K = covLINone(hyp, x, z, i) % Linear covariance function with a single hyperparameter. The covariance % function is parameterized as: % % k(x^p,x^q) = (x^p'*x^q + 1)/t2; % % where the P matrix is t2 times the unit matrix. The second term plays the % role of the bias. The hyperparameter is: % % hyp = [ log(sqrt(t2)) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '1'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode it2 = exp(-2*hyp); % t2 inverse % precompute inner products if dg % vector kxx K = sum(x.*x,2); else if xeqz % symmetric matrix Kxx K = x*x'; else % cross covariances Kxz K = x*z'; end end if nargin<4 % covariances K = it2*(1+K); else % derivatives if i==1 K = -2*it2*(1+K); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covSEisoU.m000644 000765 000024 00000002760 11447142675 017724 0ustar00carlstaff000000 000000 function K = covSEisoU(hyp, x, z, i) % Squared Exponential covariance function with isotropic distance measure with % unit magnitude. The covariance function is parameterized as: % % k(x^p,x^q) = exp(-(x^p - x^q)'*inv(P)*(x^p - x^q)/2) % % where the P matrix is ell^2 times the unit matrix and sf2 is the signal % variance. The hyperparameters are: % % hyp = [ log(ell) ] % % For more help on design of covariance functions, try "help covFunctions". % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '1'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode ell = exp(hyp(1)); % characteristic length scale % precompute squared distances if dg % vector kxx K = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx K = sq_dist(x'/ell); else % cross covariances Kxz K = sq_dist(x'/ell,z'/ell); end end if nargin<4 % covariances K = exp(-K/2); else % derivatives if i==1 K = exp(-K/2).*K; else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covConst.m000644 000765 000024 00000002172 11447142675 017640 0ustar00carlstaff000000 000000 function K = covConst(hyp, x, z, i) % covariance function for a constant function. The covariance function is % parameterized as: % % k(x^p,x^q) = s2; % % The scalar hyperparameter is: % % hyp = [ log(sqrt(s2)) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '1'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode s2 = exp(2*hyp); % s2 n = size(x,1); if dg % vector kxx K = s2*ones(n,1); else if xeqz % symmetric matrix Kxx K = s2*ones(n); else % cross covariances Kxz K = s2*ones(n,size(z,1)); end end if nargin>3 % derivatives if i==1 K = 2*K; else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covADD.m000644 000765 000024 00000006733 11447142675 017151 0ustar00carlstaff000000 000000 function K = covADD(cov, hyp, x, z, i) % Additive covariance function using a 1d base covariance function % cov(x^p,x^q;hyp) with individual hyperparameters hyp. % % k(x^p,x^q) = \sum_{r \in R} sf_r \sum_{|I|=r} % \prod_{i \in I} cov(x^p_i,x^q_i;hyp_i) % % hyp = [ hyp_1 % hyp_2 % ... % hyp_D % log(sf_R(1)) % ... % log(sf_R(end)) ] % % where hyp_d are the parameters of the 1d covariance function which are shared % over the different values of R(1) to R(end). % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. R = cov{1}; nh = eval(feval(cov{2})); % number of hypers per individual covariance nr = numel(R); % number of different degrees of interaction if nargin<3 % report number of hyper parameters K = ['D*', int2str(nh), '+', int2str(nr)]; return end if nargin<4, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode [n,D] = size(x); % dimensionality sf2 = exp( 2*hyp(D*nh+(1:nr)) ); % signal variances of individual degrees Kd = Kdim(cov{2},hyp,x,z); % evaluate dimensionwise covariances K if nargin<5 % covariances EE = elsympol(Kd,max(R)); % Rth elementary symmetric polynomials K = 0; for ii=1:nr, K = K + sf2(ii)*EE(:,:,R(ii)+1); end % sf2 weighted sum else % derivatives if i <= D*nh % individual covariance function parameters j = fix(1+(i-1)/nh); % j is the dimension of the hyperparameter if dg, zj='diag'; else if xeqz, zj=[]; else zj=z(:,j); end, end dKj = feval(cov{2},hyp(nh*(j-1)+(1:nh)),x(:,j),zj,i-(j-1)*nh); % other dK=0 % the final derivative is a sum of multilinear terms, so if only one term % depends on the hyperparameter under consideration, we can factorise it % out and compute the sum with one degree less E = elsympol(Kd(:,:,[1:j-1,j+1:D]),max(R)-1); % R-1th elementary sym polyn K = 0; for ii=1:nr, K = K + sf2(ii)*E(:,:,R(ii)); end % sf2 weighted sum K = dKj.*K; elseif i <= D*nh+nr EE = elsympol(Kd,max(R)); % Rth elementary symmetric polynomials j = i-D*nh; K = 2*sf2(j)*EE(:,:,R(j)+1); % rest of the sf2 weighted sum else error('Unknown hyperparameter') end end % evaluate dimensionwise covariances K function K = Kdim(cov,hyp,x,z) [n,D] = size(x); % dimensionality nh = eval(feval(cov)); % number of hypers per individual covariance if nargin<4, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode if dg % allocate memory K = zeros(n,1,D); else if xeqz, K = zeros(n,n,D); else K = zeros(n,size(z,1),D); end end for d=1:D hyp_d = hyp(nh*(d-1)+(1:nh)); % hyperparamter of dimension d if dg K(:,:,d) = feval(cov,hyp_d,x(:,d),'diag'); else if xeqz K(:,:,d) = feval(cov,hyp_d,x(:,d)); else K(:,:,d) = feval(cov,hyp_d,x(:,d),z(:,d)); end end end gpml-matlab-v3.1-2010-09-27/cov/covFITC.m000644 000765 000024 00000003231 11504370467 017270 0ustar00carlstaff000000 000000 function [K,Kuu,Ku] = covFITC(cov, xu, hyp, x, z, i) % Covariance function to be used together with the FITC approximation. % % The function allows for more than one output argument and does not respect the % interface of a proper covariance function. In fact, it wraps a proper % covariance function such that it can be used together with infFITC.m. % Instead of outputing the full covariance, it returns cross-covariances between % the inputs x, z and the inducing inputs xu as needed by infFITC.m % % Copyright (c) by Ed Snelson, Carl Edward Rasmussen % and Hannes Nickisch, 2010-12-21. % % See also COVFUNCTIONS.M, INFFITC.M. if nargin<4, K = feval(cov{:}); return, end if nargin<5, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode if size(xu,2) ~= size(x,2) error('Dimensionality of inducing inputs must match training inputs'); end if nargin<6 % covariances if dg K = feval(cov{:},hyp,x,'diag'); else if xeqz K = feval(cov{:},hyp,x,'diag'); Kuu = feval(cov{:},hyp,xu); Ku = feval(cov{:},hyp,xu,x); else K = feval(cov{:},hyp,xu,z); end end else % derivatives if dg K = feval(cov{:},hyp,x,'diag',i); else if xeqz K = feval(cov{:},hyp,x,'diag',i); Kuu = feval(cov{:},hyp,xu,[],i); Ku = feval(cov{:},hyp,xu,x,i); else K = feval(cov{:},hyp,xu,z,i); end end endgpml-matlab-v3.1-2010-09-27/cov/covPeriodic.m000644 000765 000024 00000003015 11511075463 020275 0ustar00carlstaff000000 000000 function K = covPeriodic(hyp, x, z, i) % Stationary covariance function for a smooth periodic function, with period p: % % k(x,y) = sf2 * exp( -2*sin^2( pi*||x-y||/p )/ell^2 ) % % where the hyperparameters are: % % hyp = [ log(ell) % log(p) % log(sqrt(sf2)) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-01-05. % % See also COVFUNCTIONS.M. if nargin<2, K = '3'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode n = size(x,1); ell = exp(hyp(1)); p = exp(hyp(2)); sf2 = exp(2*hyp(3)); % precompute distances if dg % vector kxx K = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx K = sqrt(sq_dist(x')); else % cross covariances Kxz K = sqrt(sq_dist(x',z')); end end K = pi*K/p; if nargin<4 % covariances K = sin(K)/ell; K = K.*K; K = sf2*exp(-2*K); else % derivatives if i==1 K = sin(K)/ell; K = K.*K; K = 4*sf2*exp(-2*K).*K; elseif i==2 R = sin(K)/ell; K = 4*sf2/ell*exp(-2*R.*R).*R.*cos(K).*K; elseif i==3 K = sin(K)/ell; K = K.*K; K = 2*sf2*exp(-2*K); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covNNone.m000644 000765 000024 00000004016 11447142675 017566 0ustar00carlstaff000000 000000 function K = covNNone(hyp, x, z, i) % Neural network covariance function with a single parameter for the distance % measure. The covariance function is parameterized as: % % k(x^p,x^q) = sf2 * asin(x^p'*P*x^q / sqrt[(1+x^p'*P*x^p)*(1+x^q'*P*x^q)]) % % where the x^p and x^q vectors on the right hand side have an added extra bias % entry with unit value. P is ell^-2 times the unit matrix and sf2 controls the % signal variance. The hyperparameters are: % % hyp = [ log(ell) % log(sqrt(sf2) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '2'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode n = size(x,1); ell2 = exp(2*hyp(1)); sf2 = exp(2*hyp(2)); sx = 1 + sum(x.*x,2); if dg % vector kxx K = sx./(sx+ell2); else if xeqz % symmetric matrix Kxx S = 1 + x*x'; K = S./(sqrt(ell2+sx)*sqrt(ell2+sx)'); else % cross covariances Kxz S = 1 + x*z'; sz = 1 + sum(z.*z,2); K = S./(sqrt(ell2+sx)*sqrt(ell2+sz)'); end end if nargin<4 % covariances K = sf2*asin(K); else % derivatives if i==1 % lengthscale if dg V = K; else vx = sx./(ell2+sx); if xeqz V = repmat(vx/2,1,n) + repmat(vx'/2,n,1); else vz = sz./(ell2+sz); nz = size(z,1); V = repmat(vx/2,1,nz) + repmat(vz'/2,n,1); end end K = -2*sf2*(K-K.*V)./sqrt(1-K.*K); elseif i==2 % magnitude K = 2*sf2*asin(K); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covRQard.m000644 000765 000024 00000004320 11447142675 017560 0ustar00carlstaff000000 000000 function K = covRQard(hyp, x, z, i) % Rational Quadratic covariance function with Automatic Relevance Determination % (ARD) distance measure. The covariance function is parameterized as: % % k(x^p,x^q) = sf2 * [1 + (x^p - x^q)'*inv(P)*(x^p - x^q)/(2*alpha)]^(-alpha) % % where the P matrix is diagonal with ARD parameters ell_1^2,...,ell_D^2, where % D is the dimension of the input space, sf2 is the signal variance and alpha % is the shape parameter for the RQ covariance. The hyperparameters are: % % hyp = [ log(ell_1) % log(ell_2) % .. % log(ell_D) % log(sqrt(sf2)) % log(alpha) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-08-04. % % See also COVFUNCTIONS.M. if nargin<2, K = '(D+2)'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode [n,D] = size(x); ell = exp(hyp(1:D)); sf2 = exp(2*hyp(D+1)); alpha = exp(hyp(D+2)); % precompute squared distances if dg % vector kxx D2 = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx D2 = sq_dist(diag(1./ell)*x'); else % cross covariances Kxz D2 = sq_dist(diag(1./ell)*x',diag(1./ell)*z'); end end if nargin<4 % covariances K = sf2*((1+0.5*D2/alpha).^(-alpha)); else % derivatives if i<=D % length scale parameter if dg K = D2*0; else if xeqz K = sf2*(1+0.5*D2/alpha).^(-alpha-1).*sq_dist(x(:,i)'/ell(i)); else K = sf2*(1+0.5*D2/alpha).^(-alpha-1).*sq_dist(x(:,i)'/ell(i),z(:,i)'/ell(i)); end end elseif i==D+1 % magnitude parameter K = 2*sf2*((1+0.5*D2/alpha).^(-alpha)); elseif i==D+2 K = (1+0.5*D2/alpha); K = sf2*K.^(-alpha).*(0.5*D2./K - alpha*log(K)); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covSum.m000644 000765 000024 00000003742 11447142675 017322 0ustar00carlstaff000000 000000 function K = covSum(cov, hyp, x, z, i) % covSum - compose a covariance function as the sum of other covariance % functions. This function doesn't actually compute very much on its own, it % merely does some bookkeeping, and calls other covariance functions to do the % actual work. % % Copyright (c) by Carl Edward Rasmussen & Hannes Nickisch 2010-09-10. % % See also COVFUNCTIONS.M. if numel(cov)==0, error('We require at least one summand.'), end for ii = 1:numel(cov) % iterate over covariance functions f = cov(ii); if iscell(f{:}), f = f{:}; end % expand cell array if necessary j(ii) = cellstr(feval(f{:})); % collect number hypers end if nargin<3 % report number of parameters K = char(j(1)); for ii=2:length(cov), K = [K, '+', char(j(ii))]; end, return end if nargin<4, z = []; end % make sure, z exists [n,D] = size(x); v = []; % v vector indicates to which covariance parameters belong for ii = 1:length(cov), v = [v repmat(ii, 1, eval(char(j(ii))))]; end if nargin<5 % covariances K = 0; if nargin==3, z = []; end % set default for ii = 1:length(cov) % iteration over summand functions f = cov(ii); if iscell(f{:}), f = f{:}; end % expand cell array if necessary K = K + feval(f{:}, hyp(v==ii), x, z); % accumulate covariances end else % derivatives if i<=length(v) vi = v(i); % which covariance function j = sum(v(1:i)==vi); % which parameter in that covariance f = cov(vi); if iscell(f{:}), f = f{:}; end % dereference cell array if necessary K = feval(f{:}, hyp(v==vi), x, z, j); % compute derivative else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covPPiso.m000644 000765 000024 00000004601 11447142675 017603 0ustar00carlstaff000000 000000 function K = covPPiso(v, hyp, x, z, i) % Piecewise polynomial covariance function with compact support, v = 0,1,2,3. % The covariance functions are 2v times contin. diff'ble and the corresponding % processes are hence v times mean-square diffble. The covariance function is: % % k(x^p,x^q) = s2f * (1-r)_+.^j * f(r,j) % % where r is the distance sqrt((x^p-x^q)'*inv(P)*(x^p-x^q)), P is ell^2 times % the unit matrix and sf2 is the signal variance. The hyperparameters are: % % hyp = [ log(ell) % log(sqrt(sf2)) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<3, K = '2'; return; end % report number of parameters if nargin<4, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode ell = exp(hyp(1)); sf2 = exp(2*hyp(2)); if all(v~=[0,1,2,3]), error('only 0,1,2 and 3 allowed for v'), end % degree j = floor(size(x,2)/2)+v+1; % exponent switch v case 0, f = @(r,j) 1; df = @(r,j) 0; case 1, f = @(r,j) 1 + (j+1)*r; df = @(r,j) (j+1); case 2, f = @(r,j) 1 + (j+2)*r + ( j^2+ 4*j+ 3)/ 3*r.^2; df = @(r,j) (j+2) + 2*( j^2+ 4*j+ 3)/ 3*r; case 3, f = @(r,j) 1 + (j+3)*r + (6*j^2+36*j+45)/15*r.^2 ... + (j^3+9*j^2+23*j+15)/15*r.^3; df = @(r,j) (j+3) + 2*(6*j^2+36*j+45)/15*r ... + (j^3+9*j^2+23*j+15)/ 5*r.^2; end pp = @(r,j,v,f) max(1-r,0).^(j+v).*f(r,j); dpp = @(r,j,v,f) max(1-r,0).^(j+v-1).*r.*( (j+v)*f(r,j) - max(1-r,0).*df(r,j) ); % precompute squared distances if dg % vector kxx K = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx K = sqrt( sq_dist(x'/ell) ); else % cross covariances Kxz K = sqrt( sq_dist(x'/ell,z'/ell) ); end end if nargin<5 % covariances K = sf2*pp( K, j, v, f ); else % derivatives if i==1 K = sf2*dpp( K, j, v, f ); elseif i==2 K = 2*sf2*pp( K, j, v, f ); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covMask.m000644 000765 000024 00000004012 11447142675 017440 0ustar00carlstaff000000 000000 function K = covMask(cov, hyp, x, z, i) % Apply a covariance function to a subset of the dimensions only. The subset can % either be specified by a 0/1 mask by a boolean mask or by an index set. % % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls another covariance function to do the actual work. % % The function was suggested by Iain Murray, 2010-02-18 and is based on an % earlier implementation of his dating back to 2009-06-16. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. mask = fix(cov{1}(:)); % either a binary mask or an index set cov = cov(2); % covariance function to be masked nh_string = feval(cov{:}); % number of hyperparameters of the full covariance if max(mask)<2 && length(mask)>1, mask = find(mask); end % convert 1/0->index D = length(mask); % masked dimension if nargin<3, K = num2str(eval(nh_string)); return, end % number of parameters if nargin<4, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode if eval(nh_string)~=length(hyp) % check hyperparameters error('number of hyperparameters does not match size of masked data') end if nargin<5 % covariances if dg K = feval(cov{:}, hyp, x(:,mask), 'diag'); else if xeqz K = feval(cov{:}, hyp, x(:,mask)); else K = feval(cov{:}, hyp, x(:,mask), z(:,mask)); end end else % derivatives if i <= eval(nh_string) if dg K = feval(cov{:}, hyp, x(:,mask), 'diag', i); else if xeqz K = feval(cov{:}, hyp, x(:,mask), [], i); else K = feval(cov{:}, hyp, x(:,mask), z(:,mask), i); end end else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covScale.m000644 000765 000024 00000002067 11447142675 017604 0ustar00carlstaff000000 000000 function K = covScale(cov, hyp, x, z, i) % meanScale - compose a mean function as a scaled version of another one. % % k(x^p,x^q) = sf^2 * k_0(x^p,x^q) % % The hyperparameter is: % % hyp = [ log(sf) ] % % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls other mean function to do the actual work. % % Copyright (c) by Carl Edward Rasmussen & Hannes Nickisch 2010-09-10. % % See also MEANFUNCTIONS.M. if nargin<3 % report number of parameters K = [feval(cov{:}),'+1']; return end if nargin<4, z = []; end % make sure, z exists [n,D] = size(x); sf2 = exp(2*hyp(1)); % signal variance if nargin<5 % covariances K = sf2*feval(cov{:},hyp(2:end),x,z); else % derivatives if i==1 K = 2*sf2*feval(cov{:},hyp(2:end),x,z); else K = sf2*feval(cov{:},hyp(2:end),x,z,i-1); end endgpml-matlab-v3.1-2010-09-27/cov/covRQiso.m000644 000765 000024 00000003505 11447142675 017610 0ustar00carlstaff000000 000000 function K = covRQiso(hyp, x, z, i) % Rational Quadratic covariance function with isotropic distance measure. The % covariance function is parameterized as: % % k(x^p,x^q) = sf2 * [1 + (x^p - x^q)'*inv(P)*(x^p - x^q)/(2*alpha)]^(-alpha) % % where the P matrix is ell^2 times the unit matrix, sf2 is the signal % variance and alpha is the shape parameter for the RQ covariance. The % hyperparameters are: % % hyp = [ log(ell) % log(sqrt(sf2)) % log(alpha) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '3'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode ell = exp(hyp(1)); sf2 = exp(2*hyp(2)); alpha = exp(hyp(3)); % precompute squared distances if dg % vector kxx D2 = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx D2 = sq_dist(x'/ell); else % cross covariances Kxz D2 = sq_dist(x'/ell,z'/ell); end end if nargin<4 % covariances K = sf2*((1+0.5*D2/alpha).^(-alpha)); else % derivatives if i==1 % length scale parameter K = sf2*(1+0.5*D2/alpha).^(-alpha-1).*D2; elseif i==2 % magnitude parameter K = 2*sf2*((1+0.5*D2/alpha).^(-alpha)); elseif i==3 K = (1+0.5*D2/alpha); K = sf2*K.^(-alpha).*(0.5*D2./K - alpha*log(K)); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covNoise.m000644 000765 000024 00000002706 11447142675 017632 0ustar00carlstaff000000 000000 function K = covNoise(hyp, x, z, i) % Independent covariance function, ie "white noise", with specified variance. % The covariance function is specified as: % % k(x^p,x^q) = s2 * \delta(p,q) % % where s2 is the noise variance and \delta(p,q) is a Kronecker delta function % which is 1 iff p=q and zero otherwise. The hyperparameter is % % hyp = [ log(sqrt(s2)) ] % % For more help on design of covariance functions, try "help covFunctions". % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '1'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode n = size(x,1); s2 = exp(2*hyp); % noise variance % precompute raw if dg % vector kxx K = ones(n,1); else if xeqz % symmetric matrix Kxx K = eye(n); else % cross covariances Kxz K = zeros(n,size(z,1)); end end if nargin<4 % covariances K = s2*K; else % derivatives if i==1 K = 2*s2*K; else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covSEiso.m000644 000765 000024 00000003167 11447142675 017601 0ustar00carlstaff000000 000000 function K = covSEiso(hyp, x, z, i) % Squared Exponential covariance function with isotropic distance measure. The % covariance function is parameterized as: % % k(x^p,x^q) = sf^2 * exp(-(x^p - x^q)'*inv(P)*(x^p - x^q)/2) % % where the P matrix is ell^2 times the unit matrix and sf^2 is the signal % variance. The hyperparameters are: % % hyp = [ log(ell) % log(sf) ] % % For more help on design of covariance functions, try "help covFunctions". % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '2'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode ell = exp(hyp(1)); % characteristic length scale sf2 = exp(2*hyp(2)); % signal variance % precompute squared distances if dg % vector kxx K = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx K = sq_dist(x'/ell); else % cross covariances Kxz K = sq_dist(x'/ell,z'/ell); end end if nargin<4 % covariances K = sf2*exp(-K/2); else % derivatives if i==1 K = sf2*exp(-K/2).*K; elseif i==2 K = 2*sf2*exp(-K/2); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covProd.m000644 000765 000024 00000004247 11447142675 017463 0ustar00carlstaff000000 000000 function K = covProd(cov, hyp, x, z, i) % covProd - compose a covariance function as the product of other covariance % functions. This function doesn't actually compute very much on its own, it % merely does some bookkeeping, and calls other covariance functions to do the % actual work. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if numel(cov)==0, error('We require at least one factor.'), end for ii = 1:numel(cov) % iterate over covariance functions f = cov(ii); if iscell(f{:}), f = f{:}; end % expand cell array if necessary j(ii) = cellstr(feval(f{:})); % collect number hypers end if nargin<3 % report number of parameters K = char(j(1)); for ii=2:length(cov), K = [K, '+', char(j(ii))]; end, return end if nargin<4, z = []; end % make sure, z exists [n,D] = size(x); v = []; % v vector indicates to which covariance parameters belong for ii = 1:length(cov), v = [v repmat(ii, 1, eval(char(j(ii))))]; end if nargin<5 % covariances K = 1; if nargin==3, z = []; end % set default for ii = 1:length(cov) % iteration over factor functions f = cov(ii); if iscell(f{:}), f = f{:}; end % expand cell array if necessary K = K .* feval(f{:}, hyp(v==ii), x, z); % accumulate covariances end else % derivatives if i<=length(v) K = 1; vi = v(i); % which covariance function j = sum(v(1:i)==vi); % which parameter in that covariance for ii = 1:length(cov) % iteration over factor functions f = cov(ii); if iscell(f{:}), f = f{:}; end % expand cell if necessary if ii==vi K = K .* feval(f{:}, hyp(v==ii), x, z, j); % accumulate covariances else K = K .* feval(f{:}, hyp(v==ii), x, z); % accumulate covariances end end else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covSEard.m000644 000765 000024 00000003754 11447142675 017557 0ustar00carlstaff000000 000000 function K = covSEard(hyp, x, z, i) % Squared Exponential covariance function with Automatic Relevance Detemination % (ARD) distance measure. The covariance function is parameterized as: % % k(x^p,x^q) = sf2 * exp(-(x^p - x^q)'*inv(P)*(x^p - x^q)/2) % % where the P matrix is diagonal with ARD parameters ell_1^2,...,ell_D^2, where % D is the dimension of the input space and sf2 is the signal variance. The % hyperparameters are: % % hyp = [ log(ell_1) % log(ell_2) % . % log(ell_D) % log(sqrt(sf2)) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<2, K = '(D+1)'; return; end % report number of parameters if nargin<3, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode [n,D] = size(x); ell = exp(hyp(1:D)); % characteristic length scale sf2 = exp(2*hyp(D+1)); % signal variance % precompute squared distances if dg % vector kxx K = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx K = sq_dist(diag(1./ell)*x'); else % cross covariances Kxz K = sq_dist(diag(1./ell)*x',diag(1./ell)*z'); end end K = sf2*exp(-K/2); % covariance if nargin>3 % derivatives if i<=D % length scale parameters if dg K = K*0; else if xeqz K = K.*sq_dist(x(:,i)'/ell(i)); else K = K.*sq_dist(x(:,i)'/ell(i),z(:,i)'/ell(i)); end end elseif i==D+1 % magnitude parameter K = 2*K; else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/cov/covMaterniso.m000644 000765 000024 00000003731 11447142675 020515 0ustar00carlstaff000000 000000 function K = covMaterniso(d, hyp, x, z, i) % Matern covariance function with nu = d/2 and isotropic distance measure. For % d=1 the function is also known as the exponential covariance function or the % Ornstein-Uhlenbeck covariance in 1d. The covariance function is: % % k(x^p,x^q) = s2f * f( sqrt(d)*r ) * exp(-sqrt(d)*r) % % with f(t)=1 for d=1, f(t)=1+t for d=3 and f(t)=1+t+t²/3 for d=5. % Here r is the distance sqrt((x^p-x^q)'*inv(P)*(x^p-x^q)), P is ell times % the unit matrix and sf2 is the signal variance. The hyperparameters are: % % hyp = [ log(ell) % log(sqrt(sf2)) ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-10. % % See also COVFUNCTIONS.M. if nargin<3, K = '2'; return; end % report number of parameters if nargin<4, z = []; end % make sure, z exists xeqz = numel(z)==0; dg = strcmp(z,'diag') && numel(z)>0; % determine mode ell = exp(hyp(1)); sf2 = exp(2*hyp(2)); if all(d~=[1,3,5]), error('only 1, 3 and 5 allowed for d'), end % degree switch d case 1, f = @(t) 1; df = @(t) 1; case 3, f = @(t) 1 + t; df = @(t) t; case 5, f = @(t) 1 + t.*(1+t/3); df = @(t) t.*(1+t)/3; end m = @(t,f) f(t).*exp(-t); dm = @(t,f) df(t).*t.*exp(-t); % precompute distances if dg % vector kxx K = zeros(size(x,1),1); else if xeqz % symmetric matrix Kxx K = sqrt( sq_dist(sqrt(d)*x'/ell) ); else % cross covariances Kxz K = sqrt( sq_dist(sqrt(d)*x'/ell,sqrt(d)*z'/ell) ); end end if nargin<5 % covariances K = sf2*m(K,f); else % derivatives if i==1 K = sf2*dm(K,f); elseif i==2 K = 2*sf2*m(K,f); else error('Unknown hyperparameter') end endgpml-matlab-v3.1-2010-09-27/doc/index.html000644 000765 000024 00000100354 11527514533 017633 0ustar00carlstaff000000 000000 Documentation for GPML Matlab Code

Documentation for GPML Matlab Code version 3.1

What?

The code provided here originally demonstrated the main algorithms from Rasmussen and Williams: Gaussian Processes for Machine Learning. It has since grown to allow more likelihood functions, further inference methods and a flexible framework for specifying GPs. It does not currently contain inference methods for sparse approximations. Other GP packages can be found here.

The code is written by Carl Edward Rasmussen and Hannes Nickisch; it runs on both Octave 3.2.x and Matlab® 7.x. The code is based on previous versions written by Carl Edward Rasmussen and Chris Williams.

Download, Install and Documentation

All the code including demonstrations and html documentation can be downloaded in a tar or zip archive file.

Minor changes and incremental bugfixes are documented in the changelog.

Please read the copyright notice.

After unpacking the tar or zip file you will find 6 subdirectories: cov, doc, inf, lik, mean and util. It is not necessary to install anything to get started, just run the startup.m script to set your path.

Details about the directory contents and on how to compile mex files can be found in the README. The getting started guide is the remainder of the html file you are currently reading (also available at http://gaussianprocess.org/gpml/code/matlab/doc). A Developer's Guide containing technical documentation is found in manual.pdf, but for the casual user, the guide is below.

Theory

Gaussian Processes (GPs) can conveniently be used for Bayesian supervised learning, such as regression and classification. In its simplest form, GP inference can be implemented in a few lines of code. However, in practice, things typically get a little more complicated: you might want to use complicated covariance functions and mean functions, learn good values for hyperparameters, use non-Gaussian likelihood functions (rendering exact inference intractable), use approximate inference algorithms, or combinations of many or all of the above. This is what the GPML software package does.

Before going straight to the examples, just a brief note about the organization of the package. There are four types of objects which you need to know about:

Gaussian Process
A Gaussian Process is fully specified by a mean function and a covariance function. These functions are specified separately, and consist of a specification of a functional form as well as a set of parameters called hyperparameters, see below.
Mean functions
Several mean functions are available, all start with the four letters mean and reside in the mean directory. An overview is provided by the meanFunctions help function (type help meanFunctions to get help), and an example is the meanLinear function.
Covariance functions
There are many covariance functions available, all start with the three letters cov and reside in the cov directory. An overview is provided by the covFunctions help function (type help covFunctions to get help), and an example is the covSEard "Squared Exponential with Automatic Relevance Determination" covariance function.
For both mean functions and covariance functions, two types exist: simple and composite. Whereas simple types are specified by the function name (or function pointer), composite functions join together several components using cell arrays. Composite functions can be composed of other composite functions, allowing for very flexible and interesting structures. Examples are given below and in the usageMean and usageCov functions.
Hyperparameters
GPs are typically specified using mean and covariance functions which have free parameters called hyperparameters. Also likelihood functions may have such parameters. These are encoded in a struct with the fields mean, cov and lik (some of which may be empty). When specifying hyperparameters, it is important that the number of elements in each of these struct fields, precisely match the number of parameters expected by the mean function, the covariance function and the likelihood functions respectively (otherwise an error will result). Hyperparameters whose natural domain is positive are represented by their logarithms.
Likelihood Functions
The likelihood function specifies the probability of the observations given the latent function, i.e. the GP (and the hyperparameters). All likelihood functions begin with the three letters lik and reside in the lik directory. An overview is provided by the likFunctions help function (type help likFunctions to get help). Some examples are likGauss the Gaussian likelihood and likLogistic the logistic function used in classification (a.k.a. logistic regression).
Inference Methods
The inference methods specify how to compute with the model, i.e. how to infer the (approximate) posterior process, how to find hyperparameters, evaluate the log marginal likelihood and how to make predictions. Inference methods all begin with the three letters inf and reside in the inf directory. An overview is provided by the infMethods help file (type help infMethods to get help). Some examples are infExact for exact inference or infEP for the Expectation Propagation algorithm. Further usage examples are provided for both regression and classification. However, not all combinations of likelihood function and inference method are possible (e.g. you cannot do exact inference with a Laplace likelihood). An exhaustive compatibility matrix between likelihoods (rows) and inference methods (columns) is given in the table below:
Likelihood \ Inference
 		  
GPML name		 Exact
infExact		 FITC
infFITC		 EP
infEP		 Laplace
infLaplace		 Variational Bayes
infVB		 LOO
infLOO		 type, output domain
 		 alternate names
 
Gaussian likGauss $\checkmark$ $\checkmark$ $\checkmark$ $\checkmark$ $\checkmark$ $\checkmark$ regression, IR  
Sech-squared likSech2     $\checkmark$ $\checkmark$ $\checkmark$ $\checkmark$ regression, IR  
Laplacian likLaplace     $\checkmark$   $\checkmark$ $\checkmark$ regression, IR double exponential
Student's t likT       $\checkmark$ $\checkmark$ $\checkmark$ regression, IR  
Error function likErf     $\checkmark$ $\checkmark$ $\checkmark$ $\checkmark$ classification, ±1 probit regression
Logistic function likLogistic     $\checkmark$ $\checkmark$ $\checkmark$ $\checkmark$ classification, ±1 logistic regression
logit regression

All of the objects described above are written in a modular way, so you can add functionality if you feel constrained despite the considerable flexibility provided. Details about how to do this are provided in the developer documentation.

Practice

Using the GPML package is simple. There is only one single function to call: gp, it does posterior inference, learns hyperparameters, computes the marginal likelihood and makes predictions. Generally, the gp function takes the following arguments: a hyperparameter struct, an inference method, a mean function, a covariance function, a likelihood function, training inputs, training targets, and possibly test cases. The exact computations done by the function is controlled by the number of input and output arguments in the call. Here is part of the help message for the gp function (follow the link to see the whole thing):

  function [varargout] = gp(hyp, inf, mean, cov, lik, x, y, xs, ys)

  [ ... snip ...]

  Two modes are possible: training or prediction: if no test cases are
  supplied, then the negative log marginal likelihood and its partial
  derivatives wrt the hyperparameters is computed; this mode is used to fit the
  hyperparameters. If test cases are given, then the test set predictive
  probabilities are returned. Usage:

    training: [nlZ dnlZ          ] = gp(hyp, inf, mean, cov, lik, x, y);
  prediction: [ymu ys2 fmu fs2   ] = gp(hyp, inf, mean, cov, lik, x, y, xs);
          or: [ymu ys2 fmu fs2 lp] = gp(hyp, inf, mean, cov, lik, x, y, xs, ys);

  [ .. snip ...]

Here x and y are training inputs and outputs, and xs and ys are test set inputs and outputs, nlZ is the negative log marginal likelihood and dnlZ its partial derivatives wrt the hyperparameters (which are used for training the hyperparameters). The prediction outputs are ymu and ys2 for test output mean and covariance, and fmu and fs2 are the equivalent quenteties for the corresponding latent variables. Finally, lp are the test output log probabilities.

Instead of exhaustively explaining all the possibilities, we will give two illustrative examples to give you the idea; one for regression and one for classification. You can either follow the example here on this page, or using the two scripts demoRegression and demoClassification (using the scripts, you still need to follow the explanation on this page).

Regression

You can either follow the example here on this page, or use the script demoRegression.

This is a simple example, where we first generate n=20 data points from a GP, where the inputs are scalar (so that it is easy to plot what is going on). We then use various other GPs to make inferences about the underlying function.

First, generate some data from a Gaussian process (it is not essential to understand the details of this):

  clear all, close all
 
  meanfunc = {@meanSum, {@meanLinear, @meanConst}}; hyp.mean = [0.5; 1];
  covfunc = {@covMaterniso, 3}; ell = 1/4; sf = 1; hyp.cov = log([ell; sf]);
  likfunc = @likGauss; sn = 0.1; hyp.lik = log(sn);
 
  n = 20;
  x = gpml_randn(0.3, n, 1);
  K = feval(covfunc{:}, hyp.cov, x);
  mu = feval(meanfunc{:}, hyp.mean, x);
  y = chol(K)'*gpml_randn(0.15, n, 1) + mu + exp(hyp.lik)*gpml_randn(0.2, n, 1);

  plot(x, y, '+')

Above, we first specify the mean function meanfunc, covariance function covfunc of a GP and a likelihood function, likfunc. The corresponding hyperparameters are specified in the hyp structure:

The mean function is composite, adding (using meanSum function) a linear (meanLinear) and a constant (meanConst) to get an affine function. Note, how the different components are composed using cell arrays. The hyperparameters for the mean are given in hyp.mean and consists of a single (because the input will one dimensional, i.e. D=1) slope (set to 0.5) and an off-set (set to 1). The number and the order of these hyperparameters conform to the mean function specification. You can find out how many hyperparameters a mean (or covariance or likelihood function) expects by calling it without arguments, such as feval(@meanfunc{:}). For more information on mean functions see meanFunctions and the directory mean/.

The covariance function is of the Matérn form with isotropic distance measure covMaterniso. This covariance function is also composite, as it takes a constant (related to the smoothness of the GP), which in this case is set to 3. The covariance function takes two hyperparameters, a characteristic length-scale ell and the standard deviation of the signal sf. Note, that these positive parameters are represented in hyp.cov using their logarithms. For more information on covariance functions see covFunctions and cov/.

Finally, the likelihood function is specified to be Gaussian. The standard deviation of the noise sn is set to 0.1. Again, the representation in the hyp.lik is given in terms of its logarithm. For more information about likelihood functions, see likFunctions and lik/.

Then, we generate a dataset with n=20 examples. The inputs x are drawn from a unit Gaussian (using the gpml_randn utility, which generates unit Gaussian pseudo random numbers with a specified seed). We then evaluate the covariance matrix K and the mean vector m by calling the corresponding functions with the hyperparameters and the input locations x. Finally, the targets y are computed by drawing randomly from a Gaussian with the desired covariance and mean and adding Gaussian noise with standard deviation exp(hyp.lik). The above code is a bit special because we explicitly call the mean and covariance functions (in order to generate samples from a GP); ordinarily, we would only directly call the gp function.

f0.gif

Let's ask the model to compute the (joint) negative log probability (density) nlml (also called marginal likelihood or evidence) and to generalize from the training data to other (test) inputs z:

  nlml = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y)

  z = linspace(-1.9, 1.9, 101)';
  [m s2] = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y, z);

  f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)]; 
  fill([z; flipdim(z,1)], f, [7 7 7]/8)
  hold on; plot(z, m); plot(x, y, '+')

The gp function is called with a struct of hyperparameters hyp, and inference method, in this case @infExact for exact inference and the mean, covariance and likelihood functions, as well as the inputs and outputs of the training data. With no test inputs, gp returns the negative log probability of the training data, in this example nlml=11.97.

To compute the predictions at test locations we add the test inputs z as a final argument, and gp returns the mean m variance s2 at the test location. The program is using algorithm 2.1 from the GPML book. Plotting the mean function plus/minus two standard deviations (corresponding to a 95% confidence interval):

f1.gif

Typically, we would not a priori know the values of the hyperparameters hyp, let alone the form of the mean, covariance or likelihood functions. So, let's pretend we didn't know any of this. We assume a particular structure and learn suitable hyperparameters:

  covfunc = @covSEiso; hyp2.cov = [0; 0]; hyp2.lik = log(0.1);

  hyp2 = minimize(hyp2, @gp, -100, @infExact, [], covfunc, likfunc, x, y);
  exp(hyp2.lik)
  nlml2 = gp(hyp2, @infExact, [], covfunc, likfunc, x, y)

  [m s2] = gp(hyp2, @infExact, [], covfunc, likfunc, x, y, z);
  f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)];
  fill([z; flipdim(z,1)], f, [7 7 7]/8)
  hold on; plot(z, m); plot(x, y, '+')

First, we guess that a squared exponential covariance function covSEiso may be suitable. This covariance function takes two hyperparameters: a characteristic length-scale and a signal standard deviation (magnitude). These hyperparameters are non-negative and represented by their logarithms; thus, initializing hyp2.cov to zero, correspond to unit characteristic length-scale and unit signal standard deviation. The likelihood hyperparameter in hyp2.lik is also initialized. We assume that the mean function is zero, so we simply ignore it (and when in the following we call gp, we give an empty argument for the mean function).

In the following line, we optimize over the hyperparameters, by minimizing the negative log marginal likelihood w.r.t. the hyperparameters. The third parameter in the call to minimize limits the number of function evaluations to a maximum of 100. The inferred noise standard deviation is exp(hyp2.lik)=0.15, somewhat larger than the one used to generate the data (0.1). The final negative log marginal likelihood is nlml2=14.13, showing that the joint probability (density) of the training data is about exp(14.13-11.97)=8.7 times smaller than for the setup actually generating the data. Finally, we plot the predictive distribution.

f2.gif

This plot shows clearly, that the model is indeed quite different from the generating process. This is due to the different specifications of both the mean and covariance functions. Below we'll try to do a better job, by allowing more flexibility in the specification.

Note that the confidence interval in this plot is the confidence for the distribution of the (noisy) data. If instead you want the confidence region for the underlying function, you should use the 3rd and 4th output arguments from gp as these refer to the latent process, rather than the data points.

  hyp.cov = [0; 0]; hyp.mean = [0; 0]; hyp.lik = log(0.1);
  hyp = minimize(hyp, @gp, -100, @infExact, meanfunc, covfunc, likfunc, x, y);
  [m s2] = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y, z);
 
  f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)];
  fill([z; flipdim(z,1)], f, [7 7 7]/8)
  hold on; plot(z, m); plot(x, y, '+');

Here, we have changed the specification by adding the affine mean function. All the hyperparameters are learnt by optimizing the marginal likelihood.

f3.gif

This shows that a much better fit is achieved when allowing a mean function (although the covariance function is still different from that of the generating process).

Exercises for the reader

Inference Methods
Try using Expectation Propagation instead of exact inference in the above, by exchanging @infExact with @infEP. You get exactly identical results, why?
Mean or Covariance
Try training a GP where the affine part of the function is captured by the covariance function instead of the mean function. That is, use a GP with no explicit mean function, but further additive contributions to the covariance. How would you expect the marginal likelihood to compare to the previous case?

Large scale regression

In case the number of training inputs x exceeds a few thousands, exact inference using infExact.m takes too long. We offer the FITC approximation based on a low-rank plus diagonal approximation to the exact covariance to deal with these cases. The general idea is to use inducing points u and to base the computations on cross-covariances between training, test and inducing points only.

Using the FITC approximation is very simple, we just have to wrap the covariance function covfunc into covFITC.m and call gp.m with the inference method infFITC.m as demonstrated by the following lines of code. 

nu = fix(n/2); u = linspace(-1.3,1.3,nu)';
covfuncF = {@covFITC, {covfunc}, u};
[mF s2F] = gp(hyp, @infFITC, meanfunc, covfuncF, likfunc, x, y, z);

We define equispaced inducing points u that are shown in the figure as black circles. Note that the predictive variance is overestimated outside the support of the inducing inputs. In a multivariate example where densely sampled inducing inputs are infeasible, one can simply use a random subset of the training points. 

nu = fix(n/2); iu = randperm(n); iu = iu(1:nu); u = x(iu,:);
f4.gif

Classification

You can either follow the example here on this page, or use the script demoClassification.

The difference between regression and classification isn't of fundamental nature. We can use a Gaussian process latent function in essentially the same way, it is just that the Gaussian likelihood function often used for regression is inappropriate for classification. And since exact inference is only possible for Gaussian likelihood, we also need an alternative, approximate, inference method.

Here, we will demonstrate binary classification, using two partially overlapping Gaussian sources of data in two dimensions. First we generate the data:

  clear all, close all
 
  n1 = 80; n2 = 40;                   % number of data points from each class
  S1 = eye(2); S2 = [1 0.95; 0.95 1];           % the two covariance matrices
  m1 = [0.75; 0]; m2 = [-0.75; 0];                            % the two means
 
  x1 = bsxfun(@plus, chol(S1)'*gpml_randn(0.2, 2, n1), m1);
  x2 = bsxfun(@plus, chol(S2)'*gpml_randn(0.3, 2, n2), m2);
 
  x = [x1 x2]'; y = [-ones(1,n1) ones(1,n2)]';
  plot(x1(1,:), x1(2,:), 'b+'); hold on;
  plot(x2(1,:), x2(2,:), 'r+');

120 data points are generated from two Gaussians with different means and covariances. One Gaussian is isotropic and contains 2/3 of the data (blue), the other is highly correlated and contains 1/3 of the points (red). Note, that the labels for the targets are ±1 (and not 0/1).

In the plot, we superimpose the data points with the posterior equi-probability contour lines for the probability of class two given complete information about the generating mechanism

  [t1 t2] = meshgrid(-4:0.1:4,-4:0.1:4);
  t = [t1(:) t2(:)]; n = length(t);                 % these are the test inputs
  tmm = bsxfun(@minus, t, m1');
  p1 = n1*exp(-sum(tmm*inv(S1).*tmm/2,2))/sqrt(det(S1));
  tmm = bsxfun(@minus, t, m2');
  p2 = n2*exp(-sum(tmm*inv(S2).*tmm/2,2))/sqrt(det(S2));
  contour(t1, t2, reshape(p2./(p1+p2), size(t1)), [0.1:0.1:0.9]);
f5.gif

We specify a Gaussian process model as follows: a constant mean function, with initial parameter set to 0, a squared exponential with automatic relevance determination (ARD) covariance function covSEard. This covariance function has one characteristic length-scale parameter for each dimension of the input space, and a signal magnitude parameter, for a total of 3 parameters (as the input dimension is D=2). ARD with separate length-scales for each input dimension is a very powerful tool to learn which inputs are important for predictions: if length-scales are short, inputs are very important, and when they grow very long (compared to the spread of the data), the corresponding inputs will be largely ignored. Both length-scales and the signal magnitude are initialized to 1 (and represented in the log space). Finally, the likelihood function likErf has the shape of the error-function (or cumulative Gaussian), which doesn't take any hyperparameters (so hyp.lik does not exist).

  meanfunc = @meanConst; hyp.mean = 0;
  covfunc = @covSEard; ell = 1.0; sf = 1.0; hyp.cov = log([ell ell sf]);
  likfunc = @likErf;

  hyp = minimize(hyp, @gp, -40, @infEP, meanfunc, covfunc, likfunc, x, y);
  [a b c d lp] = gp(hyp, @infEP, meanfunc, covfunc, likfunc, x, y, t, ones(n, 1));

  plot(x1(1,:), x1(2,:), 'b+'); hold on; plot(x2(1,:), x2(2,:), 'r+')
  contour(t1, t2, reshape(exp(lp), size(t1)), [0.1:0.1:0.9]);

We train the hyperparameters using minimize, to minimize the negative log marginal likelihood. We allow for 40 function evaluations, and specify that inference should be done with the Expectation Propagation (EP) inference method @infEP, and pass the usual parameters. Training is done using algorithm 3.5 and 5.2 from the gpml book. When computing test probabilities, we call gp with additional test inputs, and as the last argument a vector of targets for which the log probabilities lp should be computed. The fist four output arguments of the function are mean and variance for the targets and corresponding latent variables respectively. The test set predictions are computed using algorithm 3.6 from the gpml book. The contour plot for the predictive distribution is shown below. Note, that the predictive probability is fairly close to the probabilities of the generating process in regions of high data density. Note also, that as you move away from the data, the probability approaches 1/3, the overall class probability.

f6.gif

Examining the two ARD characteristic length-scale parameters after learning, you will find that they are fairly similar, reflecting the fact that for this data set, both inputs important.

Exercise for the reader

Inference Methods
Use the Laplace Approximation for inference @infLaplace, and compare the approximate marginal likelihood for the two methods. Which approximation is best?
Covariance Function
Try using the squared exponential with isotropic distance measure covSEiso instead of ARD distance measure covSEard. Which is best?

Acknowledgements

Innumerable colleagues have helped to improve this software. Some of these are: John Cunningham, Máté Lengyel, Joris Mooij, Iain Murray and Chris Williams. Especially Ed Snelson helped to improve the code and to include sparse approximations.

Last modified: February 15th 2011 gpml-matlab-v3.1-2010-09-27/doc/changelog000644 000765 000024 00000002105 11527514533 017503 0ustar00carlstaff000000 000000 GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox version 3.1 for GNU Octave 3.2.x and Matlab 7.x Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-15. Changes and small incremental bug fixes applied since last release. 2011-02-15, hn - added gfortran support to util/lbfgsb/Makefile (thanks to Ernst Kloppenburg) 2011-01-05 - changed parametrisation in cov/covPeriodic.m (thanks to Philipp Hennig) 2010-12-21 - fixed a stupid bug in cov/covFITC.m (thanks to Krzysztof Chalupka) 2010-12-16 - changed scalars in util/solve_chol.c to long to provide 64bit compatibility (thanks to Ernst Kloppenburg) 2010-12-14 - added infLOO to allow pseudo-likelihood optimisation instead of the marginal likelihood 2010-12-13 - fixed a performance issue in util/sq_dist.m (thanks to Krzysztof Chalupka) 2010-11-26 - fixed a bug causing numerical problems in infFITC and covFITC whenever Kuu is close to singular (thanks to Joris M. Mooij) 2010-11-24 - lbfgs compilation requires some packages not in jaunty but in hardy - updated the documentation to account for this gpml-matlab-v3.1-2010-09-27/doc/usageRegression.m000644 000765 000024 00000005201 11422307416 021146 0ustar00carlstaff000000 000000 % demonstrate usage of regression % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-07-21. clear all, close all %% SAY WHICH CODE WE WISH TO EXERCISE id = [1,1]; % use Exact % id = [1,2; 3,2; 4,2]; % compare Laplace id = [1,3; 2,3; 3,3]; % compare EP % id = [1,4; 2,4; 3,4; 4,4]; % compare VB seed = 197; randn('seed',seed), rand('seed',seed) ntr = 50; nte = 1e4; % number of training and test points xtr = 10*sort(rand(ntr,1)); % sample dataset f = @(x) sin(x)+sqrt(x); % "true" underlying function sn = 0.2; ytr = f(xtr) + randn(ntr,1)*sn; % add Gaussian noise i = randperm(ntr); nout = 3; % add outliers ytr(i(1:nout)) = 5; xte = linspace(0,10,1e4)'; % support, we test our function on cov = {@covSEiso}; sf = 1; ell = 0.4; % setup the GP hyp0.cov = log([ell;sf]); mean = {@meanSum,{@meanLinear,@meanConst}}; a = 1/5; b = 1; % m(x) = a*x+b hyp0.mean = [a;b]; lik_list = {'likGauss','likLaplace','likSech2','likT'}; % possible likelihoods inf_list = {'infExact','infLaplace','infEP','infVB'}; % allowable inference algs Ncg = 50; % number of conjugate gradient steps sdscale = 0.5; % how many sd wide should the error bars become? col = {'k',[.8,0,0],[0,.5,0],'b',[0,.75,.75],[.7,0,.5]}; % colors ymu{1} = f(xte); ys2{1} = sn^2; nlZ(1) = -Inf; for i=1:size(id,1) lik = lik_list{id(i,1)}; % setup the likelihood if strcmp(lik,'likT') nu = 4; hyp0.lik = log([nu-1;sqrt((nu-2)/nu)*sn]); else hyp0.lik = log(sn); end inf = inf_list{id(i,2)}; fprintf('OPT: %s/%s\n',lik_list{id(i,1)},inf_list{id(i,2)}) if Ncg==0 hyp = hyp0; else hyp = minimize(hyp0,'gp', -Ncg, inf, mean, cov, lik, xtr, ytr); % opt hypers end [ymu{i+1}, ys2{i+1}] = gp(hyp, inf, mean, cov, lik, xtr, ytr, xte); % predict [nlZ(i+1)] = gp(hyp, inf, mean, cov, lik, xtr, ytr); end figure, hold on for i=1:size(id,1)+1 plot(xte,ymu{i},'Color',col{i},'LineWidth',2) if i==1 leg = {'function'}; else leg{end+1} = sprintf('%s/%s -lZ=%1.2f',... lik_list{id(i-1,1)},inf_list{id(i-1,2)},nlZ(i)); end end for i=1:size(id,1)+1 ysd = sdscale*sqrt(ys2{i}); fill([xte;flipud(xte)],[ymu{i}+ysd;flipud(ymu{i}-ysd)],... col{i},'EdgeColor',col{i},'FaceAlpha',0.1,'EdgeAlpha',0.3); end for i=1:size(id,1)+1, plot(xte,ymu{i},'Color',col{i},'LineWidth',2), end plot(xtr,ytr,'k+'), plot(xtr,ytr,'ko'), legend(leg) gpml-matlab-v3.1-2010-09-27/doc/gpml_randn.m000644 000765 000024 00000002125 11422240325 020117 0ustar00carlstaff000000 000000 function x = gpml_randn(seed,varargin) % Generate pseudo-random numbers in a quick and dirty way. % The function makes sure, we obtain the same random numbers using Octave and % Matlab for the demo scripts. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-07-07. if nargin==2 sz = varargin{1}; else sz = zeros(1,nargin-1); for i=1:nargin-1 sz(i) = varargin{i}; end end n = prod(sz); N = ceil(n/2)*2; % minimal uniform random number generator for uniform deviates from [0,1] % by Park and Miller a = 7^5; m = 2^31-1; % using Schrage's algorithm q = fix(m/a); r = mod(m,a); % m = a*q+r u = zeros(N+1,1); u(1) = fix(seed*2^31); for i=2:N+1 % Schrage's algorithm for mod(a*u(i),m) u(i) = a*mod(u(i-1),q) - r*fix(u(i-1)/q); if u(i)<0, u(i) = u(i)+m; end end u = u(2:N+1)/2^31; % Box-Muller transform: Numerical Recipies, 2nd Edition, $7.2.8 % http://en.wikipedia.org/wiki/Box-Muller_transform w = sqrt(- 2*log(u(1:N/2))); % split into two groups x = [w.*cos(2*pi*u(N/2+1:N)); w.*sin(2*pi*u(N/2+1:N))]; x = reshape(x(1:n),sz); gpml-matlab-v3.1-2010-09-27/doc/usageCov.m000644 000765 000024 00000004122 11527514620 017561 0ustar00carlstaff000000 000000 % demonstrate usage of covariance functions % % See also covFunctions.m. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 clear all, close all n = 5; D = 3; x = randn(n,D); xs = randn(3,D); % create a data set % set up simple covariance functions cn = {'covNoise'}; sn = .1; hypn = log(sn); % one hyperparameter cc = {@covConst}; sf = 2; hypc = log(sf); % function handles OK cl = {@covLIN}; hypl = []; % linear is parameter-free cla = {'covLINard'}; L = rand(D,1); hypla = log(L); % linear (ARD) clo = {@covLINone}; ell = .9; hyplo = log(ell); % linear with bias cp = {@covPoly,3}; c = 2; hypp = log([c;sf]); % third order poly cga = {@covSEard}; hypga = log([L;sf]); % Gaussian with ARD cgi = {'covSEiso'}; hypgi = log([ell;sf]); % isotropic Gaussian cgu = {'covSEisoU'}; hypgu = log(ell); % isotropic Gauss no scale cra = {'covRQard'}; al = 2; hypra = log([L;sf;al]); % ration. quad. cri = {@covRQiso}; hypri = log([ell;sf;al]); % isotropic cm = {'covMaterniso',3}; hypm = log([ell;sf]); % Matern class q=3 cnn = {'covNNone'}; hypnn = log([L;sf]); % neural network cpe = {'covPeriodic'}; om = 2; hyppe = log([ell;om;sf]); % periodic ccc = {'covPPiso',2}; hypcc = hypm; % compact support poly degree 2 % set up composite covariance functions csc = {'covScale',{cgu}}; hypsc = [log(3); hypgu]; % scale by 9 csu = {'covSum',{cn,cc,cl}}; hypsu = [hypn; hypc; hypl]; % sum cpr = {@covProd,{cc,ccc}}; hyppr = [hypc; hypcc]; % product mask = [0,1,0]; % binary mask excluding all but the 2nd component cma = {'covMask',{mask,cgi{:}}}; hypma = hypgi; % additive based on SEiso using unary and pairwise interactions cad = {'covADD',{[1,2],'covSEiso'}}; % 0) specify covariance function cov = cma; hyp = hypma; % 1) query the number of parameters feval(cov{:}) % 2) evaluate the function on x feval(cov{:},hyp,x) % 3) evaluate the function on x and xs to get cross-terms kss = feval(cov{:},hyp,xs,'diag') Ks = feval(cov{:},hyp,x,xs) % 4) compute the derivatives w.r.t. to hyperparameter i i = 1; feval(cov{:},hyp,x,[],i) gpml-matlab-v3.1-2010-09-27/doc/README000644 000765 000024 00000022560 11527514533 016520 0ustar00carlstaff000000 000000 GAUSSIAN PROCESS REGRESSION AND CLASSIFICATION Toolbox version 3.1 for GNU Octave 3.2.x and Matlab 7.x Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-09-27. 0) HOW TO READ ============== If you want to get started without further delay, then read section 1) below and jump right to the examples in doc/index.html. 1) ABOUT THESE PROGRAMS ======================= This collection of matlab programs implements and demonstrates some of the algorithms described in a) the book by Rasmussen and Williams: "Gaussian Processes for Machine Learning", the MIT Press 2006, in b) the article by Nickisch and Rasmussen: "Approximations for Binary Gaussian Process Classification", JMLR 2008 and in c) the article by Candela and Rasmussen: "A Unifying View of Sparse Approximate Gaussian Process Regression", JMLR 2005. There are 6 code subdirectories: cov, doc, inf, lik, mean and util. cov/ contains covariance functions cov*.m => see covFunctions.m doc/ contains an index.html file providing documentation. This information is also available from http://www.GaussianProcess.org/gpml/code. Usage of mean, cov, classification and regression is demonstrated in usage*.m. Further details can be found in the developer documentation manual.pdf. inf/ contains the inference algorithms inf*.m => see infMethods.m lik/ contains the likelihood functions lik*.m => see likFunctions.m mean/ contains the mean functions mean*.m => see meanFunctions.m util/ contains optimisation routines, backward compatibility programs and small auxiliary scripts Before running the demos or any other part of the code, you should execute startup.m to add all necessary directories to your path. In Octave, you can also use the .octaverc file. 2) ABOUT MEX FILES ================== Some of the programs make use of the MEX facility in matlab for more efficient implementation. However, if you don't know about how to compile MEX files, you do not need to worry about this -- the code runs anyway. If you do compile the MEX files, this is automatically detected, and the program will run more efficiently. If you do not have a compiler installed on your system, you might want to consult [2]. Two components of the toolbox can be accelerated by MEX. First, there is an auxiliary functions solve_chol.m having a MEX equivalent. This can be compiled by either executing util/make.m from the Matlab/Octave command line. In addition to that, we provide a Makefile for Matlab in util/ that you can run by simply typing make in your shell after having supplied your Matlab path and your operating system. We recommend the make.m script because it works both under Matlab and Octave. Second, in order to use the L-BFGS minimiser for hyperparameter optimisation as an equivalent to minimize.m, you have to compile Peter Carbonetto's "Matlab interface for L-BFGS-B" [1]. The challenge here is the Fortran 77 code. We provide a Makefile suitable for Linux 32/64 bit and Mac whenever you have 1) g77 or 2) gfortran properly set up. Under Ubuntu, you can achieve this by installing the packages fort77 and gfortran, respectively. Compilation is done by editing util/lbfgsb/Makefile. In any case, you need to provide $MATLAB_HOME which can be found by the commands 'locate matlab' or 'find / -name "matlab"'. You can choose between two compilation modes: a) using the mex utility by Matlab [default] provide $MEX, then type 'make mex' b) without mex utility by Matlab provide $MEX_SUFFIX and $MATLAB_LIB, then type 'make nomex' In Ubuntu 10.04 LTS, the libg2c library needed for both 1)+a) and 1)+b) is not included per default. If 'ls /usr/lib/libg2c.*' does not list anything this is the case on your machine. You then whant to install the packages gcc-3.4-base and libg2c0 e.g. from http://packages.ubuntu.com/hardy/gcc-3.4-base and http://packages.ubuntu.com/hardy/libg2c0. After installation, you have to create a symbolic link by 'cd /usr/lib' and 'ln -s libg2c.so.0 libg2c.so'. [1] http://www.cs.ubc.ca/~pcarbo/lbfgsb-for-matlab.html [2] http://www.mathworks.com/support/compilers/R2010a 3) CURRENT VERSION ================== The current version of the programs is 3.0. Previous versions of the code are available at http://www.gaussianprocess.org/gpml/code/oldcode.html. 4) DIFFERENCES TO PREVIOUS VERSIONS =================================== NEW in version 3.1, 2010-10-10 ------------------------------ - following a suggestion by Ed Snelson we now support FITC regression - cov/covFITC.m and inf/infFITC.m have been added with Ed Snelson's help - the covariance interface was slightly changed to make that possible NEW in version 3.0, 2010-07-23 ------------------------------ A major code reorganisation effort did take place in the current release. First, classification and regression are now done by a single file gp.m which is completely generic in the likelihood. The previous regression program gpr.m corresponds to gp.m with Gaussian likelihood. Several other likelihoods for robust regresssion were added. Further, the code now supports mean functions with a similar specification mechanism as already used by the covariance functions. Previous implementations correspond to using meanZero.m. We merged the covariance functions covMatern3iso.m and covMatern5.iso into a single covMaterniso.m and added a covariance function for additive functions as well as the possibility to use only certain components of the data. Finally, we included covPPiso.m a piecewise polynomial covariance function with compact support. New likelihood functions in lik/: likGauss.m, likLaplace.m, likLogistic.m and likT.m New mean functions in mean/: meanConst.m, meanLinear.m, meanPow.m, meanProd.m, meanSum.m and meanZero.m New covariance functions in cov/: covADD.m, covMask.m, covPPiso.m The gprSRPP.m function which previously provided "Subset of Regressors" and the "Projected Process" approximation has now been removed. NEW in version 2.1, 2007-07-25 ------------------------------ covConst.m: fixed a bug which caused an error in the derivative of the log marginal likelihood for certain combinations of covariance functions and approximation methods. (Thanks to Antonio Eleuteri for reporting the problem) gauher.m: added the function "gauher.m" which was mistakenly missing from the previous release. This caused an error for certain combinations of approximation method and likelihood function. logistic.m: modified the approximation of moments calculation to use a mixture of cumulative Gaussian, rather than Gauss-Hermite quadrature, as the former turns out to be more accurate. NEW in version 2.0, 2007-06-25 ------------------------------ Some code restructuring has taken place for the classification code to make it more modular, to facilitate addition of new likelihood functions and approximations methods. Now, all classification is done using the binaryGP function, which (among other things) takes an approximation method and a likelihood function as an arguments. Thus, binaryGP replaces both binaryEPGP and binaryLapaceGP, although wrapper functions are still provided for backward compatibility. This gives added flexibility: now EP can also be used wth the logistic likelihood function (implemented using Gauss-Hermite quadrature). approxEP.m: New file, containing the Expectation Propagation approximation method, which was previously contained in binaryEPGP.m approxLA.m: New file, containing Laplaces approximation method, which was previously contained in binaryLaplace.m approximations.m: New file, help for the approximation methods. binaryEPGP.m: This file has been replaced by a wrapper (for backward compatibility) which calls the more general binaryGP function. binaryGP.m: New general function to do binary classification. binaryLaplaceGP.m: This file has been replaced by a wrapper (for backward compatibility) which calls the more general binaryGP function. covMatern3iso.m, covMatern5iso.m, covNNone.m, covRQard.m, covRQiso.m, cosSEard, covSEiso: now check more carefully, that persistent variables have the correct sizes, and some variable names have been modified. cumGauss.m: New file, containing code for the cumulative Gaussian likelihood function likelihoods.m: New file, help for likelihood functions logistic.m: New file, logistic likelihood NEW in version 1.3, 2006-09-08 ------------------------------ covRQard.m: bugfix: replaced x with x' and z with z' in line 36 covRQiso.m: bugfix: replaced x with x' and z with z' in line 28 minimize.m: correction: replaced "error()" with "error('')", and made a few cosmetic changes binaryEPGP.m: added the line "lml = -n*log(2);" in line 77. This change should be largely inconsequential, but occationally may save things when the covariance matrix is exceptionally badly conditioned. NEW in version 1.2, 2006-05-10 ------------------------------ added the "erfint" function to "binaryLaplaceGP.m". The erfint function was missing by mistake, preventing the use of the "logistic" likelihood. NEW in version 1.1, 2006-04-12 ------------------------------ added files: "covProd.m" and "covPeriodic.m" changes: "covSEiso.m" was changed slightly to avoid the use of persistent variables NEW in version 1.0, 2006-03-29 ------------------------------ initial version shipped with the book gpml-matlab-v3.1-2010-09-27/doc/checkmark.png000644 000765 000024 00000000306 11422240325 020255 0ustar00carlstaff000000 000000 ‰PNG  IHDR§óa*PLTE³³³¨¨¨œœœ„„„xxxlll```TTTHHH<<<$$$ ÍÍtRNS@æØfJIDAT™c````d€6 N`€8,Åà~€Àe€ˆ‹ RS!K”Áe0O€Ðâ åPÎë`ʳÁ&Õ]¹ ‡NIEND®B`‚gpml-matlab-v3.1-2010-09-27/doc/demoRegression.m000644 000765 000024 00000011500 11447142675 021000 0ustar00carlstaff000000 000000 disp(' '); disp('clear all, close all'); clear all, close all disp(' '); disp('meanfunc = {@meanSum, {@meanLinear, @meanConst}}; hyp.mean = [0.5; 1];'); meanfunc = {@meanSum, {@meanLinear, @meanConst}}; hyp.mean = [0.5; 1]; disp('covfunc = {@covMaterniso, 3}; ell = 1/4; sf = 1; hyp.cov = log([ell; sf]);'); covfunc = {@covMaterniso, 3}; ell = 1/4; sf = 1; hyp.cov = log([ell; sf]); disp('likfunc = @likGauss; sn = 0.1; hyp.lik = log(sn);'); likfunc = @likGauss; sn = 0.1; hyp.lik = log(sn); disp(' '); disp('n = 20;'); n = 20; disp('x = gpml_randn(0.3, n, 1);'); x = gpml_randn(0.3, n, 1); disp('K = feval(covfunc{:}, hyp.cov, x);'); K = feval(covfunc{:}, hyp.cov, x); disp('mu = feval(meanfunc{:}, hyp.mean, x);'); mu = feval(meanfunc{:}, hyp.mean, x); disp('y = chol(K)''*gpml_randn(0.15, n, 1) + mu + exp(hyp.lik)*gpml_randn(0.2, n, 1);'); y = chol(K)'*gpml_randn(0.15, n, 1) + mu + exp(hyp.lik)*gpml_randn(0.2, n, 1); figure(1) set(gca, 'FontSize', 24) disp(' '); disp('plot(x, y, ''+'')'); plot(x, y, '+', 'MarkerSize', 12) axis([-1.9 1.9 -0.9 3.9]) grid on xlabel('input, x') ylabel('output, y') %print -depsc f0.eps disp(' '); disp('Hit any key to continue...'); pause disp(' '); disp('nlml = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y)'); nlml = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y) disp(' ') disp('z = linspace(-1.9, 1.9, 101)'';'); z = linspace(-1.9, 1.9, 101)'; disp('[m s2] = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y, z);'); [m s2] = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y, z); figure(2) set(gca, 'FontSize', 24); disp(' '); disp('f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)];'); f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)]; disp('fill([z; flipdim(z,1)], f, [7 7 7]/8);'); fill([z; flipdim(z,1)], f, [7 7 7]/8); disp('hold on; plot(z, m); plot(x, y, ''+'')') hold on; plot(z, m, 'LineWidth', 2); plot(x, y, '+', 'MarkerSize', 12) axis([-1.9 1.9 -0.9 3.9]) grid on xlabel('input, x') ylabel('output, y') %print -depsc f1.eps disp(' '); disp('Hit any key to continue...'); pause disp(' '); disp('covfunc = @covSEiso; hyp2.cov = [0; 0]; hyp2.lik = log(0.1);'); covfunc = @covSEiso; hyp2.cov = [0; 0]; hyp2.lik = log(0.1); disp('hyp2 = minimize(hyp2, @gp, -100, @infExact, [], covfunc, likfunc, x, y)'); hyp2 = minimize(hyp2, @gp, -100, @infExact, [], covfunc, likfunc, x, y); disp(' '); disp('exp(hyp2.lik)'); exp(hyp2.lik) disp('nlml2 = gp(hyp2, @infExact, [], covfunc, likfunc, x, y)'); nlml2 = gp(hyp2, @infExact, [], covfunc, likfunc, x, y) disp('[m s2] = gp(hyp2, @infExact, [], covfunc, likfunc, x, y, z);'); [m s2] = gp(hyp2, @infExact, [], covfunc, likfunc, x, y, z); disp(' ') figure(3) set(gca, 'FontSize', 24) f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)]; disp('fill([z; flipdim(z,1)], f, [7 7 7]/8)'); fill([z; flipdim(z,1)], f, [7 7 7]/8) disp('hold on; plot(z, m); plot(x, y, ''+'')'); hold on; plot(z, m, 'LineWidth', 2); plot(x, y, '+', 'MarkerSize', 12) grid on xlabel('input, x') ylabel('output, y') axis([-1.9 1.9 -0.9 3.9]) %print -depsc f2.eps disp(' '); disp('Hit any key to continue...'); pause disp(' '); disp('hyp.cov = [0; 0]; hyp.mean = [0; 0]; hyp.lik = log(0.1);'); hyp.cov = [0; 0]; hyp.mean = [0; 0]; hyp.lik = log(0.1); disp('hyp = minimize(hyp, @gp, -100, @infExact, meanfunc, covfunc, likfunc, x, y);'); hyp = minimize(hyp, @gp, -100, @infExact, meanfunc, covfunc, likfunc, x, y); disp('[m s2] = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y, z);'); [m s2] = gp(hyp, @infExact, meanfunc, covfunc, likfunc, x, y, z); figure(4) set(gca, 'FontSize', 24) disp(' '); disp('f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)];'); f = [m+2*sqrt(s2); flipdim(m-2*sqrt(s2),1)]; disp('fill([z; flipdim(z,1)], f, [7 7 7]/8)'); fill([z; flipdim(z,1)], f, [7 7 7]/8) disp('hold on; plot(z, m); plot(x, y, ''+'');'); hold on; plot(z, m, 'LineWidth', 2); plot(x, y, '+', 'MarkerSize', 12) grid on xlabel('input, x') ylabel('output, y') axis([-1.9 1.9 -0.9 3.9]) %print -depsc f3.eps disp(' '); disp('Hit any key to continue...'); pause disp('large scale regression using the FITC approximation') disp('nu = fix(n/2); u = linspace(-1.3,1.3,nu)'';') nu = fix(n/2); u = linspace(-1.3,1.3,nu)'; disp('covfuncF = {@covFITC, {covfunc}, u};') covfuncF = {@covFITC, {covfunc}, u}; disp('[mF s2F] = gp(hyp, @infFITC, meanfunc, covfuncF, likfunc, x, y, z);') [mF s2F] = gp(hyp, @infFITC, meanfunc, covfuncF, likfunc, x, y, z); figure(5) set(gca, 'FontSize', 24) disp(' '); disp('f = [mF+2*sqrt(s2F); flipdim(mF-2*sqrt(s2F),1)];'); f = [mF+2*sqrt(s2F); flipdim(mF-2*sqrt(s2F),1)]; disp('fill([z; flipdim(z,1)], f, [7 7 7]/8)'); fill([z; flipdim(z,1)], f, [7 7 7]/8) disp('hold on; plot(z, mF); plot(x, y, ''+'');'); hold on; plot(z, mF, 'LineWidth', 2); plot(x, y, '+', 'MarkerSize', 12) disp('plot(u,1,''o'')') plot(u,1,'ko', 'MarkerSize', 12) grid on xlabel('input, x') ylabel('output, y') axis([-1.9 1.9 -0.9 3.9]) print -depsc f4.eps disp(' ');gpml-matlab-v3.1-2010-09-27/doc/usageClassification.m000644 000765 000024 00000005057 11422307416 021772 0ustar00carlstaff000000 000000 % demonstrate usage of classification % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-06-21. clear all, close all %% SAY WHICH CODE WE WISH TO EXERCISE % id = [1,1]; % use Gauss/Exact id = [1,2; 2,2; 3,2]; % compare Laplace % id = [1,3; 2,3; 3,3]; % compare EP % id = [1,4; 3,4]; % compare VB seed = 943; randn('seed',seed), rand('seed',seed) ntr = 50; nte = 1e4; % number of training and test points xtr = 10*sort(rand(ntr,1)); % sample dataset p = @(x) 1./(1+exp(-5*sin(x))); % "true" underlying probability ytr = 2*(p(xtr)>rand(ntr,1))-1; % draw labels +1/-1 i = randperm(ntr); nout = 3; % add outliers ytr(i(1:nout)) = -ytr(i(1:nout)); xte = linspace(0,10,1e4)'; % support, we test our function on cov = {@covSEiso}; sf = 1; ell = 0.7; % setup the GP hyp0.cov = log([ell;sf]); mean = {@meanZero}; % m(x) = 0 hyp0.mean = []; lik_list = {'likGauss','likErf','likLogistic'}; % allowable likelihoods inf_list = {'infExact','infLaplace','infEP','infVB'}; % possible inference algs Ncg = 50; % number of conjugate gradient steps sdscale = 0.5; % how many sd wide should the error bars become? col = {'k',[.8,0,0],[0,.5,0],'b',[0,.75,.75],[.7,0,.5]}; % colors ymu{1} = 2*p(xte)-1; ys2{1} = 0; for i=1:size(id,1) lik = lik_list{id(i,1)}; % setup the likelihood if strcmp(lik,'likGauss') sn = .2; hyp0.lik = log(sn); else hyp0.lik = []; end inf = inf_list{id(i,2)}; fprintf('OPT: %s/%s\n',lik_list{id(i,1)},inf_list{id(i,2)}) hyp = minimize(hyp0,'gp', -Ncg, inf, mean, cov, lik, xtr, ytr); % opt hypers [ymu{i+1}, ys2{i+1}] = gp(hyp, inf, mean, cov, lik, xtr, ytr, xte); % predict [nlZ(i+1)] = gp(hyp, inf, mean, cov, lik, xtr, ytr); end figure, hold on for i=1:size(id,1)+1 plot(xte,ymu{i},'Color',col{i},'LineWidth',2) if i==1 leg = {'function'}; else leg{end+1} = sprintf('%s/%s -lZ=%1.2f',... lik_list{id(i-1,1)},inf_list{id(i-1,2)},nlZ(i)); end end for i=1:size(id,1)+1 ysd = sdscale*sqrt(ys2{i}); fill([xte;flipud(xte)],[ymu{i}+ysd;flipud(ymu{i}-ysd)],... col{i},'EdgeColor',col{i},'FaceAlpha',0.1,'EdgeAlpha',0.3); end for i=1:size(id,1)+1, plot(xte,ymu{i},'Color',col{i},'LineWidth',2), end plot(xtr,ytr,'k+'), plot(xtr,ytr,'ko'), legend(leg) gpml-matlab-v3.1-2010-09-27/doc/style.css000644 000765 000024 00000000115 11422240325 017467 0ustar00carlstaff000000 000000 body {font-family: sans-serif; font-size: 16px} table {font-size: inherit;} gpml-matlab-v3.1-2010-09-27/doc/usageMean.m000644 000765 000024 00000002406 11527514620 017715 0ustar00carlstaff000000 000000 % demonstrate usage of mean functions % % See also meanFunctions.m. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 clear all, close all n = 5; D = 2; x = randn(n,D); % create a random data set % set up simple mean functions m0 = {'meanZero'}; hyp0 = []; % no hyperparameters are needed m1 = {'meanOne'}; hyp1 = []; % no hyperparameters are needed mc = {@meanConst}; hypc = 2; % also function handles are possible ml = {@meanLinear}; hypl = [2;3]; % m(x) = 2*x1 + 3*x2 % set up composite mean functions msc = {'meanScale',{m1}}; hypsc = [3; hyp1]; % scale by 3 msu = {'meanSum',{m0,mc,ml}}; hypsu = [hyp0; hypc; hypl]; % sum mpr = {@meanProd,{mc,ml}}; hyppr = [hypc; hypl]; % product mpo = {'meanPow',{3,msu}}; hyppo = hypsu; % third power mask = [0,1,0]; % binary mask excluding all but the 2nd component mma = {'meanMask',{mask,mpo{:}}}; hypma = hyppo; % 0) specify mean function % mean = m0; hyp = hyp0; % mean = msu; hyp = hypsu; % mean = mpr; hyp = hyppr; mean = mpo; hyp = hyppo; % 1) query the number of parameters feval(mean{:}) % 2) evaluate the function on x feval(mean{:},hyp,x) % 3) compute the derivatives w.r.t. to hyperparameter i i = 2; feval(mean{:},hyp,x,i) gpml-matlab-v3.1-2010-09-27/doc/manual.pdf000644 000765 000024 00001575101 11527514617 017620 0ustar00carlstaff000000 000000 %PDF-1.4 %ÐÔÅØ 5 0 obj << /S /GoTo /D (section.1) >> endobj 8 0 obj (Gaussian Process Training and Prediction) endobj 9 0 obj << /S /GoTo /D (section.2) >> endobj 12 0 obj (The gp Function) endobj 13 0 obj << /S /GoTo /D (section.3) >> endobj 16 0 obj (Inference Methods) endobj 17 0 obj << /S /GoTo /D (subsection.3.1) >> endobj 20 0 obj (Exact Inference ) endobj 21 0 obj << /S /GoTo /D (subsection.3.2) >> endobj 24 0 obj (Laplace's Approximation) endobj 25 0 obj << /S /GoTo /D (subsection.3.3) >> endobj 28 0 obj (Expectation Propagation) endobj 29 0 obj << /S /GoTo /D (subsection.3.4) >> endobj 32 0 obj (Variational Bayes) endobj 33 0 obj << /S /GoTo /D (section.4) >> endobj 36 0 obj (Likelihood Functions) endobj 37 0 obj << /S /GoTo /D (subsection.4.1) >> endobj 40 0 obj (Prediction) endobj 41 0 obj << /S /GoTo /D (subsection.4.2) >> endobj 44 0 obj (Interface) endobj 45 0 obj << /S /GoTo /D (subsection.4.3) >> endobj 48 0 obj (Implemented Likelihood Functions) endobj 49 0 obj << /S /GoTo /D (subsection.4.4) >> endobj 52 0 obj (Compatibility Between Likelihoods and Inference Methods) endobj 53 0 obj << /S /GoTo /D (subsection.4.5) >> endobj 56 0 obj (Gaussian Likelihood) endobj 57 0 obj << /S /GoTo /D (subsubsection.4.5.1) >> endobj 60 0 obj (Exact Inference) endobj 61 0 obj << /S /GoTo /D (subsubsection.4.5.2) >> endobj 64 0 obj (Laplace's Approximation) endobj 65 0 obj << /S /GoTo /D (subsubsection.4.5.3) >> endobj 68 0 obj (Expectation Propagation) endobj 69 0 obj << /S /GoTo /D (subsubsection.4.5.4) >> endobj 72 0 obj (Variational Bayes) endobj 73 0 obj << /S /GoTo /D (subsection.4.6) >> endobj 76 0 obj (Laplace Likelihood) endobj 77 0 obj << /S /GoTo /D (subsubsection.4.6.1) >> endobj 80 0 obj (Laplace's Approximation) endobj 81 0 obj << /S /GoTo /D (subsubsection.4.6.2) >> endobj 84 0 obj (Expectation Propagation) endobj 85 0 obj << /S /GoTo /D (subsubsection.4.6.3) >> endobj 88 0 obj (Variational Bayes) endobj 89 0 obj << /S /GoTo /D (subsection.4.7) >> endobj 92 0 obj (Student's t Likelihood) endobj 93 0 obj << /S /GoTo /D (subsubsection.4.7.1) >> endobj 96 0 obj (Laplace's Approximation) endobj 97 0 obj << /S /GoTo /D (subsection.4.8) >> endobj 100 0 obj (Cumulative Logistic Likelihood) endobj 101 0 obj << /S /GoTo /D (subsubsection.4.8.1) >> endobj 104 0 obj (Laplace's Approximation) endobj 105 0 obj << /S /GoTo /D (section.5) >> endobj 108 0 obj (Mean Functions ) endobj 109 0 obj << /S /GoTo /D (subsection.5.1) >> endobj 112 0 obj (Interface) endobj 113 0 obj << /S /GoTo /D (subsection.5.2) >> endobj 116 0 obj (Implemented Mean Functions) endobj 117 0 obj << /S /GoTo /D (subsection.5.3) >> endobj 120 0 obj (Usage of Implemented Mean Functions) endobj 121 0 obj << /S /GoTo /D (section.6) >> endobj 124 0 obj (Covariance Functions) endobj 125 0 obj << /S /GoTo /D (subsection.6.1) >> endobj 128 0 obj (Interface) endobj 129 0 obj << /S /GoTo /D (subsection.6.2) >> endobj 132 0 obj (Implemented Covariance Functions) endobj 133 0 obj << /S /GoTo /D (subsection.6.3) >> endobj 136 0 obj (Usage of Implemented Covariance Functions) endobj 137 0 obj << /S /GoTo /D [138 0 R /Fit ] >> endobj 143 0 obj << /Length 1375 /Filter /FlateDecode >> stream xÚ…WKoÛ8¾÷WèTÈ@ÍRÔ{om‘f»HvƒÖÀ¶{ %Ú&*K†H'ñ¿ßiK‰ƒ½˜œ?Í›4¶nßñ7ÖÏ«w¿æe””L$y­6QR,yTÔ5Ky­ÚèŸxµS‹W|üšÎtÓ”eEHNéöáþŽ´æˆ)Á9¯µZÔU< Ýzx^,EYÅj4zè‰HYrHXç!–eɪºŠ–iÎ`q@_äØÑ¡›öIŽ-í¿K³?£<Þ{Z~—}¯ íÿÔÍ/mš~'**Æ €åà³ÇýªÖãQŽ'ÒNª´3žÆ›a$þ½lvºWĽSrìu¿½’ᤪç"€C½fiÛò0ßV7ÆKç‰]vQ|î Ç&/—ʸ€úž9{ÈUO¡ñÓs½}:@šžõÞ‡o@ç n>ëþÜg— õr‹Ä—N‚è'¸ØP])røj]§“àÁ?îﾟ»—WŒ\J9g¢Ê¼KÅ¥Á–©HbhTÕoíŽ(WÈ°Ú ÞûP˜@u“†;í9Úz˜›ªg½Ö¶§E-Ц4±Á"Ô 25lÔ³U½™€âÍÊ,X&^|z×y“d¬Î²y~·;8)’2–†V áhlw"R[/îºáÉ«˜ƒj¦¡vLŒ®. ¸œÔh÷WncŽÛ­Bà^¿®â½’žƒN.¯5˜?Tû‚u7ã¡dp¼  Û>ùÕ }ۅ苺&—#¸ j¯]g[½q#ÌŸ› 4”ºÔ¡¯>\³Ù±›Ò¤pÀU±-¦5¯!Ýx8ætJ¢®«ØÜ<ƒeç± èŠƒÜN8õnÜÔcÄXíôˆ®ÇwÎâèõF¼r™àð:ýKuz7 hRV]BŒdéÝÉól:šA ×;yè¤ó++±› Õ§@ù™ ªëÕ²¡#ÕË9?íš»Ýî†Ö|Túçâò’Y—³âC’b›i‘Îôhà_›Ð'® Á¨1×ßH®¤Åªf×ëFvDîe¤}AÎMT[(§"‘0y¬ÔîÂGOAîi’åEümCÜÓpôÍ:*âôƒ}{Ÿ”¥ºÚÈ=””‰zÒîâÊ(“¯Œµ`eQþÿƒ1/Œx{åÕvvåp¡&U8q4Ê›bûÖ­æðóûÕ·?֘ćš9{2*w½AÑÉRu ×ÊZ5ú§JÉOŠy ž¤_ØÖ-F7ÆÊѪ–ÍÞäô¿ƒ³š× ½°!øðOË”q9Ìù߀›Õ»ÿ££ endstream endobj 138 0 obj << /Type /Page /Contents 143 0 R /Resources 142 0 R /MediaBox [0 0 612 792] /Parent 151 0 R /Annots [ 139 0 R 140 0 R 141 0 R ] >> endobj 139 0 obj << /Type /Annot /Border[0 0 1]/H/I/C[0 1 1] /Rect [144.994 542.879 341.928 554.764] /Subtype/Link/A<> >> endobj 140 0 obj << /Type /Annot /Border[0 0 1]/H/I/C[0 1 1] /Rect [152.075 530.999 415.497 542.809] /Subtype/Link/A<> >> endobj 141 0 obj << /Type /Annot /Border[0 0 1]/H/I/C[0 1 1] /Rect [272.178 411.372 366.877 423.258] /Subtype/Link/A<> >> endobj 144 0 obj << /D [138 0 R /XYZ 71 791.866 null] >> endobj 145 0 obj << /D [138 0 R /XYZ 72 754.004 null] >> endobj 142 0 obj << /Font << /F57 146 0 R /F37 147 0 R /F59 148 0 R /F55 149 0 R /F61 150 0 R >> /ProcSet [ /PDF /Text ] >> endobj 187 0 obj << /Length 901 /Filter /FlateDecode >> stream xÚí™ÏOÛ0Çïü¹-=ÄóïØÇ14ie´CÖº%Z›DMÊà¿ŸSÛ%)iÒI h›S׉퟿ïù=èM=è]ž@û{:<ù|Á¤‡( ”co8ñBì…”H™7{wþYš*)òÁÏáWÓ %*ûB/Àp"MW4$¥þe´Ìó8J…ÿm‘ŽTž›‡á@bÅIœLMK”Œ]?5ŽGEœ–ïÄ©OÊ!í” 36c ï•™Õ ÀúUßiÖ4ér­0t}.–‰´‹"¤\ÿGWåz|Áëó fWÉD-T2Rf7ª¸Oǹù˜}Í~Lع„Õ¹”êcìˆf‰P‚DÐ?ŒF…ùpe®Á `îá•òµ¿©±®íÙ» “Ì—õVhøRécø:ÊfÑH}„Ô·ú%Ëéc<Œ%ŽŽ—$$ÒG°%q¶š©Qa©ÙsŸESÛÀ`ØclÅH ÆÁýh¯¸E3ƒò4zRÚ<™$ýÁß Õš7E?@N­†K§ï+ ¿Ž«Y|Ÿ¦Ö9— ‘DònÒn÷Åê>R'ÝUoÆÑ ö»î)n=4ÔÉø•ŽP“¨tŽ”±ÃdýVÈIr+÷Wól¦æ:.Töд"AÉAJ:í@e%ý,gZÎų¸x²‚®Š?J%›äòxKp0.þïrÛVÍ:VÍ̪뗀ªu„‚ô'n;PÌ@é`ª uåWB¾åJÀäÅûm,y—q®\†æØùãôÀ40â ¯-¾§H¬°5°Z£xÈŽü*´1¨*Þ3L^»¢¿t#{U¿\±ƒ7áΛ|X|_h/6wn£WÁ]`‘^;1b¸“ †F¿˱¾lUjñRĬ©K¸h» VØö2ØJ-htÂÞj—óåLÓyp®9Æyî{»f‹½ÅÕ©áö¤-E€jË«Þ˜IÚÞ(—[xÎ4­ªnLº®b²¥ê†d¼¾=Ìånûtâëmp{:‘­3¸›éĦíý•Ï·:ÍÙEæÀÙ<ìmM­§›Ú‰e=‡x8 Ú ?VC¸Ñ³ô¡ O× ÖZái·…ù®…Þ‹Çëïl{rŠo–}åø kXt ú¹h¡ÇÙw4Ïb@„R „€»’p= 9žü=b„A endstream endobj 186 0 obj << /Type /Page /Contents 187 0 R /Resources 185 0 R /MediaBox [0 0 612 792] /Parent 151 0 R /Annots [ 152 0 R 153 0 R 154 0 R 155 0 R 156 0 R 157 0 R 158 0 R 159 0 R 160 0 R 161 0 R 162 0 R 163 0 R 164 0 R 165 0 R 166 0 R 167 0 R 168 0 R 169 0 R 170 0 R 171 0 R 172 0 R 173 0 R 174 0 R 175 0 R 176 0 R 177 0 R 178 0 R 179 0 R 180 0 R 181 0 R 182 0 R 183 0 R 184 0 R ] >> endobj 152 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [71.004 714.988 284.785 727.579] /A << /S /GoTo /D (section.1) >> >> endobj 153 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [71.004 690.528 167.321 703.12] /A << /S /GoTo /D (section.2) >> >> endobj 154 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [71.004 668.395 177.232 678.662] /A << /S /GoTo /D (section.3) >> >> endobj 155 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 654.835 190.182 665.107] /A << /S /GoTo /D (subsection.3.1) >> >> endobj 156 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 638.908 233.043 651.558] /A << /S /GoTo /D (subsection.3.2) >> >> endobj 157 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 625.358 231.451 637.573] /A << /S /GoTo /D (subsection.3.3) >> >> endobj 158 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 611.809 196.563 624.46] /A << /S /GoTo /D (subsection.3.4) >> >> endobj 159 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [71.004 589.74 189.407 600.007] /A << /S /GoTo /D (section.4) >> >> endobj 160 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 576.18 162.342 586.452] /A << /S /GoTo /D (subsection.4.1) >> >> endobj 161 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 562.631 155.633 572.903] /A << /S /GoTo /D (subsection.4.2) >> >> endobj 162 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 546.703 278.709 559.354] /A << /S /GoTo /D (subsection.4.3) >> >> endobj 163 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 533.154 392.043 545.805] /A << /S /GoTo /D (subsection.4.4) >> >> endobj 164 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 521.983 211.465 532.256] /A << /S /GoTo /D (subsection.4.5) >> >> endobj 165 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 508.434 222.058 518.706] /A << /S /GoTo /D (subsubsection.4.5.1) >> >> endobj 166 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 492.507 267.952 505.157] /A << /S /GoTo /D (subsubsection.4.5.2) >> >> endobj 167 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 478.957 266.36 491.172] /A << /S /GoTo /D (subsubsection.4.5.3) >> >> endobj 168 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 465.408 231.472 478.059] /A << /S /GoTo /D (subsubsection.4.5.4) >> >> endobj 169 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 451.859 204.778 464.51] /A << /S /GoTo /D (subsection.4.6) >> >> endobj 170 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 438.31 267.952 450.96] /A << /S /GoTo /D (subsubsection.4.6.1) >> >> endobj 171 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 424.761 266.36 436.975] /A << /S /GoTo /D (subsubsection.4.6.2) >> >> endobj 172 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 411.211 231.472 423.862] /A << /S /GoTo /D (subsubsection.4.6.3) >> >> endobj 173 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 400.04 217.312 410.313] /A << /S /GoTo /D (subsection.4.7) >> >> endobj 174 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 384.113 267.952 396.764] /A << /S /GoTo /D (subsubsection.4.7.1) >> >> endobj 175 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 370.564 262.956 383.214] /A << /S /GoTo /D (subsection.4.8) >> >> endobj 176 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [112.458 357.015 267.952 369.665] /A << /S /GoTo /D (subsubsection.4.8.1) >> >> endobj 177 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [71.004 334.945 168.778 344.776] /A << /S /GoTo /D (section.5) >> >> endobj 178 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 321.385 155.633 331.658] /A << /S /GoTo /D (subsection.5.1) >> >> endobj 179 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 305.458 255.047 318.109] /A << /S /GoTo /D (subsection.5.2) >> >> endobj 180 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 291.909 298.683 304.559] /A << /S /GoTo /D (subsection.5.3) >> >> endobj 181 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [71.004 269.84 191.807 279.67] /A << /S /GoTo /D (section.6) >> >> endobj 182 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 256.279 155.633 266.552] /A << /S /GoTo /D (subsection.6.1) >> >> endobj 183 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 240.352 281.109 253.003] /A << /S /GoTo /D (subsection.6.2) >> >> endobj 184 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [87.367 226.803 324.745 239.454] /A << /S /GoTo /D (subsection.6.3) >> >> endobj 188 0 obj << /D [186 0 R /XYZ 125.795 791.866 null] >> endobj 189 0 obj << /D [186 0 R /XYZ 72 731.826 null] >> endobj 185 0 obj << /Font << /F59 148 0 R /F18 190 0 R /F57 146 0 R >> /ProcSet [ /PDF /Text ] >> endobj 199 0 obj << /Length 3186 /Filter /FlateDecode >> stream xÚ½Ërã¸ñ¾_¡[䊅€ @•ÚC^»™ì#{pe™( ¶™¥H-IÙ3ùút£¤!gR•‹ÕšF¿0ßÀøõx±çØÊ-˲-ÃO‚’÷-="c™Î‰Ô»î¶ëŽàfzìAè+ ¬r–‹`¤@3â%ðáðd ¹¥‘~kú `GÞ<ô§óe2› €÷\ñú|úÍ©ž 3Z˜µ€Ç¾%›- Æ•\è›ÄUiJ¸Ð™‡zjž,šÞ¶ýMŸêá¡!CQÛübÚæ±Çcá˜Bà7ÓH3ç|À£µ'KÕ­>ccNGcôq`îñã¨Ôv—R9®A¦fÑæ•Þ>a¡Ì¬\áçÔMKàx6‡æ=çâP;IÃ$ž¾a†9ßþä>3µ§C^Šáì© [eJ±ôÚÏmL‚I²MÒÊÕ¬u§>åB¸èçXOÍ„!«T±yâÂÉ™'}f£¹ßR‡qäl ´T¼™Á w4ÏËSN°ö‘ «Áüzip/™qäçfÆI°¨§æhÃ.Z•"Á><Rä 1…ºm¦4ebéõ÷­›6¦¡Nq¹"þÛ¸ì"ÄÂhu«Ò¦Ñó¹mõ¾5¯G @fbÆN¦²‚öÿ+Zìl˜¯¾‡ôÏç< ¦S’ÑG$ìdãƒ7 )™Œ‰vèÛÖešrN  55Æ¡ÔJr—0Рؼn‡ªXyu!aôQpç#¢p:©ØGpa¶\ h¤ˆ"[*<>²‘W$Ä{!DnÜoM8N" “xn& ¦Þ Œ1"˜–Âûºö¨@A\zÙ§j a÷rX·)BH¨’@JI2X“$Vª*P‡<òKVC«²lfÎYmM}pR°V¿>Ëh­‹<‡O%zŸ8wRJgœ0•N Ä·¦5'( FG‹Ê‹]VJ&±8ëB”^ì48]ln¨~°ŠAÔ†ßa[0 á<]xOàEÌŸIDXÎœCAºÄp!Ź 0‡ìMÛ?c·1úÞÅâO ƒÇÚš®§_Ý`}ÔF«É—1.l™«Q‚÷.‚™ÓÙG·~ðĺùEPûœ¸å4óý*&~ âj˸bŠ¯C˜\86Ž£òFæ³V(.I$Ö™'‚(TEKK’ùKÊÝÒ¥|k‚ºdB¬ÂE!%£ÊJ¨| FtiKlã\á›C( -•]dB ³A«Ø~C_D ·‹B9÷¹Ó}Åè—U¹5Õ¹ Z¥.X8š³ñ%—e­òv5XTÇÂ8†žÃÚ¡½PzƵ8Sà¸3ØhÖÃG"Ib¨‚Û%ëç‹.ÇšzÀÎô–†¡îÀüûüÁxx›e–Ø¥-íR _ש85ἊºTX™À‚5¡à…‚C`üîÊcÍïÑÜ×—ÖÅßf\Uÿ1C¿¬¿ <üS\!}F˜Õ/ìÐrYoi×[°J'P”ƒ¤WNQãdlÔ¸4yÒ¶Ê{e Fö[rV¬#ì;k|z`}SÅ’¢ÒÞÞ}˨ôÚÞuÒÞEì½pö-%ì]ÇöžÊò e)œ(ˆÆqT¥êR“6,ž0BNÀt>€pì8~áR¸š¶|'àp v+‹ŠýÇ‘_ÄÍÐ" N›Þbíeˆ|q’‹Êw·ÔÆMÏnƒüÖ·ÌÖhÒÞ3œz=aØSQƒàôš] ãÚ‘™o%e{å%\.E`¾îHùst…è.Œ_4€LÛë%š´rå¸V@óeM •Î|ÅxïpäÕ§ Æg [bß'‰å¬yd :×ۿ߃SÞbã fªÝºÄkÒ¡?!ä"ó "œ},XÛ¤à%“³gœSžQ0™‡ÞãUŠÓÍ×a Zb¨Y>ÖdUxW+—låLTÊ#ÿ˜"W25ûëk\ÉW%¿ÊÕ¿“"¯ ';¥´U±ªÊW6´æ£Táæá»tø†b!ÄF%Ós¹n£$1çÕ²Zž¯®)—ÅÏØ€OTWÄ¥œm3S€`bÜ –û5m)Å|Á2ëDUŸ§“bÀ¬“?§vÊ2ðÛjiMJ]³&ó嬩ÌÖÖ¤áo•åž$¿bQ˜Ý.ÂüÌJÎEê˜Óstú:™Ú2¨MôÊ“ËþÍl®¡Q«ÖwAØp—¥½›Æ_òHœpý‰ï!¶ÿ*«y5vx{3v¿¹Ö`¢ [Õ­(ÖûþÉXßaÿ¸œü·¢\B­ô–¥d\y㌫¡øš)yò`Eµòâ·x•õÓÝMvµ`€ÎAÕ«¥vÿWÒÀÙçŽççDú¶9¾„BµúDúÎ^-: aàµR òj±ðù›­/ ÀXîƯ…hĪÞ¢ Hd!›Û›U%¶—î2š£öó|×Ñ‚ÿo²[wÅI³ã¹Æðäþ‘Ê?LÂ:Ý^ {)–ÿR%ld§§U%Y¦Ä¢òOƒÈÑíò®Ïÿßýs”Y5Cøp8ßmÏo¹6…ÜÏÈ)éIè`”|[ÊTToA„î‰Ùñy´w-+Æêƒû÷£ù‰tyã… ge†]JiK)K9_àüåî«ÿõî½ endstream endobj 198 0 obj << /Type /Page /Contents 199 0 R /Resources 197 0 R /MediaBox [0 0 612 792] /Parent 151 0 R /Annots [ 191 0 R 192 0 R 193 0 R 194 0 R 195 0 R 196 0 R ] >> endobj 191 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [419.766 709.253 427.824 721.904] /A << /S /GoTo /D (section.2) >> >> endobj 192 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [457.335 597.134 465.393 609.435] /A << /S /GoTo /D (section.3) >> >> endobj 193 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [217.665 429.811 225.723 442.461] /A << /S /GoTo /D (section.4) >> >> endobj 194 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [507.705 380.197 515.763 392.847] /A << /S /GoTo /D (section.5) >> >> endobj 195 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [295.894 303.833 303.952 316.135] /A << /S /GoTo /D (section.6) >> >> endobj 196 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [218.184 272.311 226.242 284.961] /A << /S /GoTo /D (section.3) >> >> endobj 200 0 obj << /D [198 0 R /XYZ 71 791.866 null] >> endobj 6 0 obj << /D [198 0 R /XYZ 72 754.004 null] >> endobj 205 0 obj << /D [198 0 R /XYZ 99.944 210.638 null] >> endobj 197 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F18 190 0 R /F70 201 0 R /F74 202 0 R /F79 203 0 R /F80 204 0 R /F81 206 0 R /F83 207 0 R >> /ProcSet [ /PDF /Text ] >> endobj 219 0 obj << /Length 3805 /Filter /FlateDecode >> stream xÚí]I㸾ϯð%€ˆ5ÜIe0—A2I.ƒFå2ËÁeË.£½ÁVU俇z¤dJ–«dûÉe÷Ð*-\ÄÇï­|"½Iô>~Cüïß|û£LzTÄ\(Ö{÷4ëiÁc"dïaÔû5bú”=<¥~øç·?Þ.iL¨´•ÁÍ“•»§Z%5±,oùñy1̦˅¿ÑVFâ„$4¿±¯’X*ÙëssbÜeÃÔ„÷23¢÷®Ô—÷awϸh¹Ï´‰¦÷›mWÓá`6Ûúmkp°\§ªÏ•7÷Š£¹ñ$¶v¦`<Ü>K¨é)‰Öޱϯ'=wð Þ‰ êÈ;bJ²è =ÛRE 8B™A9…óK8^TšéýÚW’D¿ú“1íŵê?Á-ZD/ððJ;¯ÂÇ:šÀ™eÙ‰¼áZ3¶Žß?ô'Ñ÷ùÉŸá2ZÙXô‘äC_j;a}Î"Kœæ—ý5¿[æC°§pÙÖ$îBÍ¡šÊ”ö¦DÏ á‰%”/»çd4ƒS®ÚÏåÆ¢?ìuÿм'þÛq{…©È. ]¶¾…7p—D{ ðÓW¡Vyg@n•ù–lŸ®w7ðØh,ø†Åù”ÎV Ä®-žíHI<–ľ«ÜiZÒzëQ&:¶×•Qr¼QNÓl:˜Mÿ3ÈǸišÊ—kÙq‰84#c¦Œšò¯shãt.†ià%)L %l玔’1/åçKK&€EXãZ`FI à ÙÖŽ'PN¡Ü/#r8± Ñ–‚DºÀ1™Wì€ÃOÃæ®ÚºÌD©÷õ”ªrh80ô“ªÜΫ[¢T'%0–æ-7Pf(U+œGÀr¨Z@3"Úà4@æ“œ:l™âT*rnãÙYÁ)_Þ9‚ãÊQ¥É–`Ÿ¯‰Ê²R-ÖDmi1‹¤ÄéG6ž É‘•¤Ž“è¿–ù 1êã/!|ÕóM™_ö<Ö¾œÐϲBK}­£ÿÁ­ßµ&¹¦HKR$PR(WP.¡tçq€Å°~‹Šì­öΠÖ(y‚Ó…{6X@1€9[Àœ°æ¬i ‹2*(£W*xɼw˜JO¾¼ÆJ§Ìkb’a¢µœ4\ÎWÏ™×9³t“¹£Õ:M‡‡¤BJ´íU‰¤ Q,¤Âv¨ª\˜? ”³`)mT7jÔ–+Zsh/°}0‰í ã–2¼½à§,Gå$ZŽ­€¹qÇãÁp:³¢tZü¿Ì™«"Q¶LìŠ_Lü‰§ÔÝò´]¥ëÕ`=˜§YYÏrì~îçãÏîw¾¥3Åßù%Í\Å™pÆÜ­¾£Ï‹Mš‚àgÙ§<–")dd;õÞN1ð¿£4Lg©·7LëV‰ñr6[~á Q; ·2·¦5 Äã „ÜQN Īš!vu$ ªZÅ.‘è=³Dݾ!ìñÎÀ‘ëßbp”Ë‹ÆфxûÄ®MÙV/Iô1ˆœYdÇ›R¨Ó^ËX 4©Iôs©´8 R$¬F]ó ‰ iêGàL=ã’ÍÒ_Z^5Þ”"²¿ƒpt–ð]°tÇÞGå<¹T{“)ë`ZD1ŽÉ£÷¢ê!Yi C’%NµžA«’ôõÊÓkNEr¤yÅäöž€åfmV¾*•›K¨z,¯*$ºÓÌ7êŽóI¾”³Yè—Ÿ±–ûÖÏÖÒËìðC0Žm ¼!â+u Â&aè&·ÂÚë[Í \ÎËTŽÞÀ ïì¶|©Zí ºäƯ §¢ ¿T¡U«˜¯aŒó3txvK¼´l -q§„‚KÇ›ˆÁU£!IN}`e´åÊ ”)”îê¦ìa5M$ó'^ }Û®Ä[É;ù¬$µ¯³… ¥€—¶Ô…®+à5Ù“)CrßÉ‘hÖ…iÈýcÅÿ’{^nQK\ `9çAËÌ`¸Û•ÎÜ7Ž‡ÁYp~ e[ A^³† Œ ˜qã0>yRygÕK¹´ö­œ=¿€ë%”ô70+­AÛ`Ù¬ ÈP… ` å´ÔÌŠóHXáV€Ê‰¯fˆps0ñW‘Ô·$WP¦¥lqh‘š–‹ôT“×aN¯AsüJ§düµñoÎ÷¯ó-°ªîD7ìú¥~Th(ƒ@z¶ª)c™TrŽÃQ9Ž}áOðÂÑ4ÌYðb?RUxþ)°Å-½Õî*_x+ºEeEüCè<[ž‡sê@ÜÀ»é–Ž™¯»Ç0`ïãÀRÐ7qs­›P¨jp$î'%¤9¨‰ä¼ÜÛ†ï¼V]¼ é-¡u2é€ *Yü›Ã˜/ÛbAÌSÀo_ê'÷°Ø ›ÛšH!/öü*Ð’Üw<îÌà|ùÔ^ÈÄ©‚™¾ÀÔ¬ïî YÅÚ‘ Œ-Í(;¡nÞl-åÄ(³lpgùÔn$×Al¥TöÅ›A‘à{ð„~h%Ù…Vâɒ܋ˆ÷c][Nzþ(dërH—‹ ‹ oPDbNADãé«šÏÉìùmôžóÜá<ÇyÕÜàFD܈hÑ]D´Ó^9P—è""ºëš"¢[©Š9«GòÔÕƒ‹¤ÊOÆ3”« œüÔM[K ƒR‚dߪõ¢i¤¾bÏ=–¥¥A°@½P³~ÁôV󢜽€®öBÞ«è=ð¬ÉÝßl1U¥®’ É²Ú7ü'^ƒÛ`¹¬VåÜfè,°|»ó³ó`Ê[C(½˜:œîÌ•É«ÁéËêĬ¼‰ï"^wPgvi·3ïM«%Ù©Ý5ºT»%ñgjj·$ÞrXí>c";õ†ïÐwUÙ^´ ¹µã¿ë‚ O½ ‡¤\Ï#âÑj#%¶Íù¹²U£­DÔv¿ ž°Í|`Á|PVv%ëW†4ù:UH(lßÊO˜ÎQq™m¬Ý)mȶûBòXr§iC 6 ƒ·öKR°T#Ù±j„‡ŸtFL—¸[{ϬÐ6`W×$N¸åW#x¯Ds%Èýě̂ã,Ø?ŠºCT¾b9c²×|•r×:Ðw•¾b¬äÿPÒ9¾'1Ûu…ÔŒŒ'S¥À¥Ê“û!ßu¼õ†HOJëÀ˜š¼CcÊä¶y¾KKôæòð픿`öÈó¬±ñ—îgEÛÎïXoð{,¹­ýµüê2:µÓñÁhÆ!×gm›@K+àsK1ŸýbÜ®ùË‚l´z+ÕVXƒ Hï`—'‹Z9áÂS Þ5±£Ñ­ém3Ƕe“Ìå 3©œÉ›Ô•Hn´óm¿ïÀˆ¶À‚\ôÐ3–‘~ëƒóJV¥ã.z:FíéÛ"ŒNôŸ‘YîR$^©«‰/Ü|sºÇ[`–xƒ¼fÉðšÜ^š?'ùîû–¦y§Àbq鱩o@ƒ”Lš#¢™Ü ]œ)K°šCðœWL¦ >ó¼RL¦ŒßÓ1~O™ÏØžÇݹôKß¹ôKãjÑEJîv\‘·ï$J¼¹Cwr%^w·½B€·±ðÔ®ßÅÁÈuL £·àƒÐ‰ñn\'O­ºo“Ko\*Òÿº$2ÏPÎòŒŸ±-ë%È8<ôY†Ã¼ÃH‡1·œ+¶‹5)·ŽÕbMÊ dµX“2*%ÛÛ‚¶Á!jv+D= ¢uÇ— ê2‡\ó®idíBKÆ‚MÀ&Øl*›€· ×¥CC‹ϺØûþvnMSy½)VÑy€'ž´¡¿•4 ²¾\„-è ÑŽ#Ž»;ÙKdfð÷ †‰‘Y(9r „äµY©]œæ®úÇ@&œ“úå8é…‰Û !{b8E¼ý¹‡D³®× †ÆQ“ß(u`}© ¯y}LOòVÐ|VD|[4ÿ|yˆVrE4çp¶t9$‚m‡Ä n‚Œleœ:³‘ïØ[`™{AÁÚb…üüë‹6ó !d¤…té"ÿÀJvƒšÒ,y5èPu°ç^%•ü™«@ßw/ I9Õ·²þ¶—X ¹Œ•EVu+V¹Ky±#ÌIEóCK“py C(ržyÜJ¿Îöö%O‚,œYÃ:N†sr?"Îá\@ ”ùœ,›?*[NnŒ-#’ì2Jý*Lö[X¶’[YsÛË­9Æ×$ÛYÐ뮘°,¾ÞºÀâéËÙE«.Èrù•ø]ùÍø]ËMS‚\bg¸¨ÖAú¡ó·1¶¦"ûÖ ÊçÖ­âëPW#»§}ú Ó> &ìVüõÍ’üÂÍÈ7ª;#C‡òÁ–oŽOÅuÆJß[^梁ª§ÀJ:H¥à(¼Ï/tÜZÿÐW=Mh|2ŒÜ‚Óßh¦º`Îá§÷˜FÕà„ºØÓÐ\O]|?ÁR-~8Þ:;¢Î µÎV™Âž 1??fê9š^O¾Ÿ£i/nÃ*ÂÅ­¨#¸[gÿta –bŽµ¡]x3BCŠêÒ²9fÓ½FH%ow»ï­¾`†‚O{¡¶êE›ç  ¤A2µECµ“y[èv±ñwt»£ÛݺE7MöBxÃ[Ö¿ý½ÿ}å“QOÝ êï2Ýõî¨w)Ô›íÅ„žÍ[‘éôm¡Û]¦»£Ûݾt{?™îf"Ûg~¹G/ÕwŽÀ;Ö5Þuêd¾¼ó9Ž\g”lñ¹¾·³¶ŸŒn"È–nö²¥»cs [zâ;”žúqØ&ÔCÙt¿Ð6%qBš?É…Ö½>ç1õž‘ªcäßüM$òž endstream endobj 218 0 obj << /Type /Page /Contents 219 0 R /Resources 217 0 R /MediaBox [0 0 612 792] /Parent 151 0 R /Annots [ 208 0 R 209 0 R 210 0 R 211 0 R 212 0 R 213 0 R 214 0 R 215 0 R 216 0 R ] >> endobj 208 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.935 692.124 51.396 700.563] /A << /S /GoTo /D (noweb.NW2i2LRH-4XnUPC-1) >> >> endobj 209 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [101.92 689.487 113.381 702.383] /A << /S /GoTo /D (noweb.NW2i2LRH-4XnUPC-1) >> >> endobj 210 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [159.469 665.813 171.432 677.917] /A << /S /GoTo /D (noweb.NW2i2LRH-KwRQ2-1) >> >> endobj 211 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [147.285 653.858 159.248 665.962] /A << /S /GoTo /D (noweb.NW2i2LRH-1uBDkH-1) >> >> endobj 212 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [127.888 641.902 139.349 654.007] /A << /S /GoTo /D (noweb.NW2i2LRH-2o3xLS-1) >> >> endobj 213 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [205.823 594.082 217.786 606.186] /A << /S /GoTo /D (noweb.NW2i2LRH-OTtxn-1) >> >> endobj 214 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 527.954 51.396 536.752] /A << /S /GoTo /D (noweb.NW2i2LRH-KwRQ2-1) >> >> endobj 215 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [155.276 525.316 167.239 538.381] /A << /S /GoTo /D (noweb.NW2i2LRH-KwRQ2-1) >> >> endobj 216 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [417.204 525.316 428.665 538.381] /A << /S /GoTo /D (noweb.NW2i2LRH-4XnUPC-1) >> >> endobj 220 0 obj << /D [218 0 R /XYZ 125.795 791.866 null] >> endobj 10 0 obj << /D [218 0 R /XYZ 72 754.004 null] >> endobj 222 0 obj << /D [218 0 R /XYZ 81.963 693.21 null] >> endobj 224 0 obj << /D [218 0 R /XYZ 72 690.483 null] >> endobj 225 0 obj << /D [218 0 R /XYZ 83.504 693.21 null] >> endobj 226 0 obj << /D [218 0 R /XYZ 83.504 681.255 null] >> endobj 227 0 obj << /D [218 0 R /XYZ 83.504 669.3 null] >> endobj 228 0 obj << /D [218 0 R /XYZ 83.504 657.345 null] >> endobj 229 0 obj << /D [218 0 R /XYZ 83.504 645.389 null] >> endobj 230 0 obj << /D [218 0 R /XYZ 83.504 633.434 null] >> endobj 231 0 obj << /D [218 0 R /XYZ 83.504 621.479 null] >> endobj 232 0 obj << /D [218 0 R /XYZ 83.504 609.524 null] >> endobj 233 0 obj << /D [218 0 R /XYZ 83.504 597.569 null] >> endobj 234 0 obj << /D [218 0 R /XYZ 81.963 529.039 null] >> endobj 235 0 obj << /D [218 0 R /XYZ 72 526.312 null] >> endobj 236 0 obj << /D [218 0 R /XYZ 83.504 529.039 null] >> endobj 237 0 obj << /D [218 0 R /XYZ 83.504 517.084 null] >> endobj 238 0 obj << /D [218 0 R /XYZ 83.504 505.129 null] >> endobj 239 0 obj << /D [218 0 R /XYZ 83.504 493.174 null] >> endobj 240 0 obj << /D [218 0 R /XYZ 83.504 481.219 null] >> endobj 241 0 obj << /D [218 0 R /XYZ 83.504 469.264 null] >> endobj 242 0 obj << /D [218 0 R /XYZ 83.504 457.308 null] >> endobj 243 0 obj << /D [218 0 R /XYZ 83.504 445.353 null] >> endobj 244 0 obj << /D [218 0 R /XYZ 83.504 433.398 null] >> endobj 245 0 obj << /D [218 0 R /XYZ 83.504 421.443 null] >> endobj 246 0 obj << /D [218 0 R /XYZ 83.504 409.488 null] >> endobj 247 0 obj << /D [218 0 R /XYZ 83.504 397.533 null] >> endobj 248 0 obj << /D [218 0 R /XYZ 83.504 385.577 null] >> endobj 249 0 obj << /D [218 0 R /XYZ 83.504 373.622 null] >> endobj 250 0 obj << /D [218 0 R /XYZ 83.504 361.667 null] >> endobj 251 0 obj << /D [218 0 R /XYZ 83.504 349.712 null] >> endobj 252 0 obj << /D [218 0 R /XYZ 83.504 337.757 null] >> endobj 253 0 obj << /D [218 0 R /XYZ 83.504 325.802 null] >> endobj 254 0 obj << /D [218 0 R /XYZ 83.504 313.846 null] >> endobj 255 0 obj << /D [218 0 R /XYZ 83.504 301.891 null] >> endobj 256 0 obj << /D [218 0 R /XYZ 83.504 289.936 null] >> endobj 257 0 obj << /D [218 0 R /XYZ 83.504 277.981 null] >> endobj 258 0 obj << /D [218 0 R /XYZ 83.504 266.026 null] >> endobj 259 0 obj << /D [218 0 R /XYZ 83.504 254.071 null] >> endobj 260 0 obj << /D [218 0 R /XYZ 83.504 242.115 null] >> endobj 261 0 obj << /D [218 0 R /XYZ 83.504 230.16 null] >> endobj 262 0 obj << /D [218 0 R /XYZ 83.504 218.205 null] >> endobj 263 0 obj << /D [218 0 R /XYZ 83.504 206.25 null] >> endobj 264 0 obj << /D [218 0 R /XYZ 83.504 194.295 null] >> endobj 265 0 obj << /D [218 0 R /XYZ 83.504 182.34 null] >> endobj 266 0 obj << /D [218 0 R /XYZ 83.504 170.384 null] >> endobj 267 0 obj << /D [218 0 R /XYZ 83.504 158.429 null] >> endobj 268 0 obj << /D [218 0 R /XYZ 83.504 146.474 null] >> endobj 269 0 obj << /D [218 0 R /XYZ 83.504 134.519 null] >> endobj 270 0 obj << /D [218 0 R /XYZ 83.504 122.564 null] >> endobj 271 0 obj << /D [218 0 R /XYZ 83.504 110.609 null] >> endobj 272 0 obj << /D [218 0 R /XYZ 83.504 98.653 null] >> endobj 273 0 obj << /D [218 0 R /XYZ 83.504 86.698 null] >> endobj 274 0 obj << /D [218 0 R /XYZ 83.504 74.743 null] >> endobj 217 0 obj << /Font << /F59 148 0 R /F87 221 0 R /F57 146 0 R /F18 190 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 307 0 obj << /Length 4254 /Filter /FlateDecode >> stream xÚí]ëã¶ÿ~…¿çnñ-¥´MsiZ´’ë—\V¦×ÆÚ²!ÉÞÛ¼þö’CJ¦dÙ–lîí^rX€æʟÙßpfèpt; G_½í'R©þ»U¹¿øû›Ÿ½fb1Ç|ôf:â" ”Ž%AHÙèÍdôvLëwoþUWŽ6 Ð(×ÿšì7_éz˜SâÁX€þУ‚K¾½½FQŽ×W×W–ß2Ž…Ö˜jM§9¤HS•r½Píp êx,!]Cê>)œ|i iâäç®lÝâ0ㄧ^ªczÑ®I<ž©>†j4>*¥dü}Èúб*­–SÏX3\¥y=oÑø=¤sH—šò¥ÿ¾­¨.!?s† —ú'€LoPƒ€Þ›WeòFÃ]̡Ů햾F(ˆYÅÐGÀx¡1é·*ö˜"扟…‰`º+n`øúñÄK3B¯­êµŸÍbK<9¤ ð0C<¤»<ßúa"vRż'J_£I€•dÀV&¾ ïMQ%¬ÍìwLQ¼bP†IH'SJk–U1VÃj=5/NR×Ùûå€g÷ƒ<='¶ã[ØgÒ£ø¶[´[ù!ÈÒžD[È¿†üÆR Î§Ž€Ÿ;å3È=E>ýJRª¶/{yå CÌëG Ìéô?5†¨6笼°¸èwžñà iUk{ù×ß8ëœÖ@:r€tE#¿ *ÀÀ)+-"q¿Û±³)ö§#~¬é€¾õ+IZužÍUé3áî¬QÍg¯i¶òªA€Ÿ½æÈ)@‚HÔn×KEëX(r^­òùí¬4¯¨nDêNá ˜Vï°Äi4”ÇU‰yk¢º?[\#LƒéÁŠ‰½¶PÄaŒ`H1Þ¤S„˜Òa£4`œtŒÒ©¶9ÌïCŒ:*Bœ"DU1­lÒÓ~d>bóˆýdæÛÿ1µŸÜ~Ƶv؇h0SÔ“&Õœ£œ…ˆû9]-j|¶þ¿MËjj‹Õ²z%›—ód1/Ý'5­j¡Ãñ=è‰j’š}—ùt•«Í‘ÙTæ2K¥ù7É&:ÃÇ™Î'öa=¥úŸR¥ÉÒfv“Q˜—“ܾ™I9‘“_Ù VÙâÁV5“¦|.‹ÍÂV¦çÅùÒí¡3+ŠõLi±>‘KÕs9 :I» èa©Šâ ¤Ø‚”›ðH—ì ªeøÑ,CäD nº¡Q,N!#Ìq@m £d¶!˜ ‹mx¶iâ0`¸7 ^ªÉXj’,ÍmŠäVvƒEk$ÈÒH¬ ‹&$î3ÂXƒíäÇþ¸ÎW©,,ZI~»YʬìZ~Š€ìV—M:€øŒÁ"¢ùt ª£Mg2½3œ=(²ãáÃDAD¢z˜òÀ¢òãzξ,Â<@m £¯§¦[†c¨L%}t~5=²>æ›zÓUžË´|eþ¿·µÉt¶²/šnÂ6%€}ªL.§Ïz²#µŠ¼ ;"{ÛÀeG¦dovÔw FjµpÔ±Š›{ðWR²’‹Wb7¹vpTã GéL§¶¹*t§jQdw'5Æ aŒ9äo![s‡[ ::þ´!”€VÀùgøçg[nhC*?·€¹£¡¿@Ýq®`¦â ìîakF1(­úÉ^·ûÊ£ŽhP«Á°uÞοl¼z¬,ÿïJÞ&€ºo!/ûÖ¡Ô Ï{•íÓŽßöî<·«½€þÎ['Xd—#sNòýTߣy©ÖÞùœÑ?û«Ìì[Ö"Îã¶w°qtðúX…`}èBPÿf¢¸¥^4J0!«æ3èÊôi:ë³Ú7­|ÏÄyäõ˜ûì 8ã<ƒÆÛÞývbÍÏ¡+wO²(ß÷¦|ì³á¸uìv¼a€š`þ8€‹¿l êøýô:Ù5íZþ8£‡C"¢Ö×,°nO$Hœq‡Àp´ÝÌ› å¶â*Æ>+ž>V§Þ{Ü‹^âèÙ ýh°ÐÇľp„>~*¡Ï> ýJa>V0Û§Z”4o–扄> ô‘_´ñð$ +2|¤W €ÏsØWú s|¿s̱s¨­ÿÎOÐs!æõ!~5íè""üŸ<<#òQœ<<ñyä¡Ø‡!˜!±ß]ðð1b {wt¢Æ?¹ãöS:¹u ê'jkot¶?Ũ¶¬Ê#îO§]o¢¸åz󭄸^9(…tSÈéfažNä4_ ýÅ6Ylda¾˜‚_‹z(—ëòÁ<«-¡¯Œ/ ˜/u™t•me^v9çDƒï,Ù´Ë=‡AmOÔµvyù 5ѵ©p1¿;UO¹²#Ùd•çÖÆUõ Y_"„z5ý]µÕ•/e’uv€lç¹Ù«ªtµíôXR²3Í–Ö¦œÊÅ¢2KçÉCe‹¶VëLjƒ$¸Šñ8PV$ýuåcbmÎÛ$Ÿ'7j úY›ŠBHËÜ<ñáýr–C‹ã‡ˆ–C :i{Ö¶Î]M£íÙøûsÐã8ð4¿ÇÁ»—ã-§ù²=ÂÒVƹ“7…'†Ï3-î®ƨ6/WE(QØÜzÿê„Ž™Ä)¤_Öñ¹•i¸#H-?Ú3NŠfÍ#/ÑKˆÄ˜ù‹®*œè*?nÒœ€n^­ãÒ¤–P¦„ûk¬¹òƒÃ#z8 œ–Ëd'—»H >ºzø©#¾U*Pƒ ÔZ4Ñ­—nîX‡žÝoÌ»ÿˆ ýÒcÔËâ£Òû<ºÅ‚ _tË­S²»Ç> ŠÂÐ)ð tK|E˜¥åž|µYdªNº‚ôÞÙ…“sð ­C;â{c‘ ]ÛeÖˆhKƒË7Ï Ï¥ÐF qd‚Mp-.¸(Ȭ hwTÝcma%ؾRÒ¹-‘eÔCÏ)¼I“ØÞObÈÏÄóo"Ì}2~nÏ~3{1‹Ç^nÅÿ¦‡üÕÊ»_xŠ‰´9‘1 +(.Ú¿¬ûÐÝ“>×SÎÆ_¤ÿ>-ÄŒmÌ¿t¢ÔdƒÃZØrªv-I¨e± €9… E’5ÂŒ,¹}çDJç¹E »O1 ü²Å¡WÌ ±üº—Œ9ùc`Îî%ö<ùÜØèÇv/胹ÞÞÀsÇ®’ìÁf^µ‰:¼wöô†‰þî¶Fšï…¦³°}å í·?@:«BÜëçH×8;‘f‡¤Ådø!&Ã&sˆe 1ø …{â"Õ4îÐw/øQ”vJ¹8w{‹ëL%…¹ضÏTŽïkMK@xQ›®l¾ÝdØÏLlbËÕµ?ÛúV5÷!qSìË6!ðQ)ìòt¡ÈKšå±eåû!«Ž .tÁ¿±,Ù£™]^™‹J4ñí»„ôa€…i2ÜSÃÂþ5,Åõ|a'ËŽ™'EÇÆP CMƒyŸ=;ˆ>hc<ê⮇tÄîTCb‹t÷Q ³Í±¨åµ¿ó¸ÄÉH÷ôå€)€–λxXØqÏð|ÖºÓz>ðÐÃÖUçÀ¦µ,€%¬ÃÚÒJ°Ó(GUò('Þ6| ôç†gsâVôíÞ¡©7vêÜ£·­Î¨+]Ù]uµ8[¸ÑaàI6Ê;P…$µ¼X§[HU·jßšŽ ($j¥Í;xYØÞŸF!aa{…„…¾ö'®’#+yòXõPÂ5‚õ¢U£î.(mçrÑßó¢-*¢–5(õTat¶ÔÛ]Í­ˆ‘}K{­çA©w -å¤qBlΆ_Cú5¤o ý¢7i¢¨GÜõvdñ!Ã/ï0üîú$ OB«§L¾Ä¿(¡(þ4æU}`N,¿ðy‰lµu¤¯KˆÖŽ†´»§©Z¯Hö ÔŸíÏ1 y²÷%NôíY|%lóôùŠ‡“„…ß×í±¨î°EÚÞ çªè;H¿r´úsWñÉ;ãùxg4WÞ íuϮ˗{F›\÷-Ì çòŒ»†{FëÝOîŸÜ3†›wž=ß;ã±ÎXÎÜ=Ê?£%®zûg<‚ÍÄ£†UÙ'ÿŒgáŸánàê::Ö1_¦®&CaGOrôÍÒµ{ò?®Ñd CÜÂ-C?v@/'rSs÷÷ Ã¿‚‚R³!yÆMž_ØkÝ´à˜%e•“&SÌÔñ: ©i ®¾kßg^ج׋9\ù¨Ë²Noy<¬½ÆU=Üå¯gu™”éìpçäûµLK}ÿã5!Ä\ÁiÊæÒfUO+ôfwÆâ­ >U¯‚C1”a\kW˜ê«Ã™3Âà@PÅÝÝÞ+¼@×v Zj« Í-œ:Æ@¯±´7ÓR¯Y)SðªuoÝD~†8«®¾3¡"ÍÊ|.[á j)ìדUöòŠãÒïçEÙ3!!´ ^ÇûÇ# »²Pí¶CÝ=èw ÒÆ>û8kHûõ8£¸¡õNkÁR ËEpsµÞ½XÕ]¬¡¯Ó´ð¤Ò13©èev™Py€t H-8â´;¸ bè¼w3ñçíÓjñO­OƒyrçË Sÿ>·³3{YËnÝ<Á®Âù¥@éü„‰±N|ZŸÍìÿR`äÉY@‹2µ¤¾H:4ÚÖ}‡‹1wT´´­/u¤ûVž1]Žè£èü'xõsH¹ÚmáÚgLSï0Ú“ ·jµQ¨—u7vïhÂofræðrÈaö’›<ϼ°Œ;H¹vJÊ¡“º& ¹ÎÃåúU¸®1ÍÜ8ÇM¹O®ŸÒ¼¨Jz¶ÌéåŠàªvÕO¶¦D­ÀzóÕ¢Ö™á¡ÉàXh¨sí¸:äŽYÕýq£ÒšUw?£Zøë 9µ­Æ_8?¤xëüò‰´øÄãrÜÃö˜;JüÌ_õèé‰i˜ž¶9b?—ðã÷!õ Í C3È#‡füfP7-O£¾x»_7Ù“¥oïy79Ö„·ðì|i_F'O°ðÔŒïy#}‘&ú¢5úâç /ê /r}©î´54vÈ{»u ×¼ô‡Q+Œã Š¢.Nðå›ÿ8u+ endstream endobj 306 0 obj << /Type /Page /Contents 307 0 R /Resources 305 0 R /MediaBox [0 0 612 792] /Parent 151 0 R /Annots [ 278 0 R 279 0 R 280 0 R 281 0 R 282 0 R 283 0 R 284 0 R 285 0 R 286 0 R 287 0 R 288 0 R 289 0 R 290 0 R 291 0 R 292 0 R 293 0 R 294 0 R 295 0 R 296 0 R 297 0 R 298 0 R 299 0 R 300 0 R 301 0 R 302 0 R 303 0 R 304 0 R ] >> endobj 278 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 679.783 165.939 691.887] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 279 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.935 665.014 51.396 673.454] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 280 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [149.811 662.377 161.272 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 281 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [326.967 662.377 338.93 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-KwRQ2-1) >> >> endobj 282 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [339.43 662.377 346.408 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-2yIl95-1) >> >> endobj 283 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [346.908 662.377 353.886 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-4ZZkMy-1) >> >> endobj 284 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [354.386 662.377 366.349 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-1oKncb-1) >> >> endobj 285 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [366.849 662.377 378.812 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 286 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [379.313 662.377 391.276 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-1KNHNk-1) >> >> endobj 287 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [391.776 662.377 403.739 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-4BH61R-1) >> >> endobj 288 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [404.239 662.377 416.202 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-4Mxm99-1) >> >> endobj 289 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 662.377 428.665 675.442] /A << /S /GoTo /D (noweb.NW2i2LRH-3dLroe-1) >> >> endobj 290 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 570.355 51.396 579.153] /A << /S /GoTo /D (noweb.NW2i2LRH-1uBDkH-1) >> >> endobj 291 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [141.934 567.717 153.898 580.782] /A << /S /GoTo /D (noweb.NW2i2LRH-1uBDkH-1) >> >> endobj 292 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [417.204 567.717 428.665 580.782] /A << /S /GoTo /D (noweb.NW2i2LRH-4XnUPC-1) >> >> endobj 293 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [161.113 555.999 172.072 568.103] /A << /S /GoTo /D (noweb.NW2i2LRH-CpvDB-1) >> >> endobj 294 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [193.781 544.043 205.744 555.994] /A << /S /GoTo /D (noweb.NW2i2LRH-1fgq0k-1) >> >> endobj 295 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [188.221 532.088 199.18 544.193] /A << /S /GoTo /D (noweb.NW2i2LRH-27nDss-1) >> >> endobj 296 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [40.437 492.432 51.396 500.872] /A << /S /GoTo /D (noweb.NW2i2LRH-CpvDB-1) >> >> endobj 297 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [157.076 489.795 168.035 502.86] /A << /S /GoTo /D (noweb.NW2i2LRH-CpvDB-1) >> >> endobj 298 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 489.795 428.665 502.86] /A << /S /GoTo /D (noweb.NW2i2LRH-1uBDkH-1) >> >> endobj 299 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 365.095 51.396 373.893] /A << /S /GoTo /D (noweb.NW2i2LRH-1fgq0k-1) >> >> endobj 300 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [192.847 362.457 204.81 375.354] /A << /S /GoTo /D (noweb.NW2i2LRH-1fgq0k-1) >> >> endobj 301 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 362.457 428.665 375.354] /A << /S /GoTo /D (noweb.NW2i2LRH-1uBDkH-1) >> >> endobj 302 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [40.437 130.161 51.396 138.601] /A << /S /GoTo /D (noweb.NW2i2LRH-27nDss-1) >> >> endobj 303 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [186.76 127.524 197.719 140.589] /A << /S /GoTo /D (noweb.NW2i2LRH-27nDss-1) >> >> endobj 304 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 127.524 428.665 140.589] /A << /S /GoTo /D (noweb.NW2i2LRH-1uBDkH-1) >> >> endobj 308 0 obj << /D [306 0 R /XYZ 71 791.866 null] >> endobj 309 0 obj << /D [306 0 R /XYZ 83.504 755.001 null] >> endobj 310 0 obj << /D [306 0 R /XYZ 83.504 743.045 null] >> endobj 311 0 obj << /D [306 0 R /XYZ 83.504 731.09 null] >> endobj 312 0 obj << /D [306 0 R /XYZ 83.504 719.135 null] >> endobj 313 0 obj << /D [306 0 R /XYZ 83.504 707.18 null] >> endobj 314 0 obj << /D [306 0 R /XYZ 83.504 695.225 null] >> endobj 315 0 obj << /D [306 0 R /XYZ 81.963 666.1 null] >> endobj 316 0 obj << /D [306 0 R /XYZ 72 663.373 null] >> endobj 317 0 obj << /D [306 0 R /XYZ 83.504 666.1 null] >> endobj 275 0 obj << /D [306 0 R /XYZ 81.963 571.441 null] >> endobj 318 0 obj << /D [306 0 R /XYZ 72 568.713 null] >> endobj 319 0 obj << /D [306 0 R /XYZ 83.504 571.441 null] >> endobj 320 0 obj << /D [306 0 R /XYZ 83.504 559.485 null] >> endobj 321 0 obj << /D [306 0 R /XYZ 83.504 547.53 null] >> endobj 322 0 obj << /D [306 0 R /XYZ 81.963 493.518 null] >> endobj 323 0 obj << /D [306 0 R /XYZ 72 490.791 null] >> endobj 324 0 obj << /D [306 0 R /XYZ 83.504 493.518 null] >> endobj 325 0 obj << /D [306 0 R /XYZ 83.504 481.563 null] >> endobj 326 0 obj << /D [306 0 R /XYZ 83.504 469.608 null] >> endobj 327 0 obj << /D [306 0 R /XYZ 83.504 457.653 null] >> endobj 328 0 obj << /D [306 0 R /XYZ 83.504 445.697 null] >> endobj 329 0 obj << /D [306 0 R /XYZ 83.504 433.742 null] >> endobj 330 0 obj << /D [306 0 R /XYZ 81.963 366.181 null] >> endobj 331 0 obj << /D [306 0 R /XYZ 72 363.454 null] >> endobj 332 0 obj << /D [306 0 R /XYZ 83.504 366.181 null] >> endobj 333 0 obj << /D [306 0 R /XYZ 83.504 354.226 null] >> endobj 334 0 obj << /D [306 0 R /XYZ 83.504 342.271 null] >> endobj 335 0 obj << /D [306 0 R /XYZ 83.504 330.315 null] >> endobj 336 0 obj << /D [306 0 R /XYZ 83.504 318.36 null] >> endobj 337 0 obj << /D [306 0 R /XYZ 83.504 306.405 null] >> endobj 338 0 obj << /D [306 0 R /XYZ 83.504 294.45 null] >> endobj 339 0 obj << /D [306 0 R /XYZ 83.504 282.495 null] >> endobj 340 0 obj << /D [306 0 R /XYZ 83.504 270.54 null] >> endobj 341 0 obj << /D [306 0 R /XYZ 83.504 258.584 null] >> endobj 342 0 obj << /D [306 0 R /XYZ 83.504 246.629 null] >> endobj 343 0 obj << /D [306 0 R /XYZ 83.504 234.674 null] >> endobj 344 0 obj << /D [306 0 R /XYZ 83.504 222.719 null] >> endobj 345 0 obj << /D [306 0 R /XYZ 83.504 210.764 null] >> endobj 346 0 obj << /D [306 0 R /XYZ 83.504 198.809 null] >> endobj 347 0 obj << /D [306 0 R /XYZ 81.963 131.247 null] >> endobj 348 0 obj << /D [306 0 R /XYZ 72 128.52 null] >> endobj 349 0 obj << /D [306 0 R /XYZ 83.504 131.247 null] >> endobj 350 0 obj << /D [306 0 R /XYZ 83.504 119.292 null] >> endobj 351 0 obj << /D [306 0 R /XYZ 83.504 107.337 null] >> endobj 352 0 obj << /D [306 0 R /XYZ 83.504 95.382 null] >> endobj 353 0 obj << /D [306 0 R /XYZ 83.504 83.427 null] >> endobj 354 0 obj << /D [306 0 R /XYZ 83.504 71.471 null] >> endobj 305 0 obj << /Font << /F57 146 0 R /F55 149 0 R /F73 223 0 R /F61 150 0 R /F18 190 0 R >> /ProcSet [ /PDF /Text ] >> endobj 371 0 obj << /Length 4260 /Filter /FlateDecode >> stream xÚí]ã¶ñý~ž¤Õ&±"’"Eåp-P4 Ò‚"8 hïò ³e¯z¶äJò^ö.Éo/9$%êöds¯ ,À¥)Š’3Ãùdp³¹ n¾zèÿH”òïÍ©}óäo/ž|ö%nb?f˜Ý¼XßDØyx…ÄBzóbuóÒ‹nxñfœàÆG7~xSÊŸªúÝWrj ƒ±2ñÞOoaè•MI¼B×íqo^.¢À{К} ƒ#ù¦þtcÈûsçÕS}‰÷­øõPî | e e9i¤)_¢\΋PoínHä-” {÷Îf¡€r2X–ÈË¡\BYC{õB?uõÙˆxwbÈØ{€r» ìC ˆaZ«EÕwVÏÊæ-Ñ^¹psÎ`*¨'Pn ¥„zªKgÛÁ¼·bH ŸÅ0C,VÊÅro€LшE1 Ì2ÎA#1,Ç=úm‚]Ö€ yŒ“õØØB1 B~L5ã=6Ü ÅÄdeÈÂPàÇ[¸ÁÁDFèME˜À2«¾qó±ø˜†~Ñp„»Z”âm£+TV™A¹¶êKÓS,`mµPæn(*€¸&è7 ùòNûÔ «\¹á$Þ­ÊÍ"+ À°¤X[^Ü@*+iîjâËf3¹f¤ÿò`µ,¡l7Ÿ7ýÁÏêr-Ã’»ukò›`¡Š©fL¶­Eµ»ÁJƒ¦nhiÝ”Ü:4#Mu¹VI|XY£ |ê`q€:¢´O€›á@ö9ª¸^¼î !€³Œ—¢t8OþC!<_f5Ž€ƒQmá*¡¾´Ì‚û=M± Û6Ö‘Ï°eDª5Rö?AZƒnß²ÏlŸ€K›(Ö”˜AùÊ/,1b=ÙRœpeH/ÏO°C?·€+7OÝñ;ݪ%u,)Áá¹uDm–§$uŒqÝ9y߈×¥$Å5ÓΠ=Ú%ÄçQ£ 3š€‹k}náæ#5œ!fÛ‚µÉWqëuy:]VptgóØPw*3ÙP—+? ÜVÔàèÓ¬RÞ$qÒ<ƒò(ÁÍļ?Ý‚H-û2è4ž'Y_è#:ͳØ+ÅÆ¢lzë"­á¤UOØ õ¤ñغõÌþËŠG9¸´w' £2¥òGï Üëvgþ|¹‚£\%48¡à\Z ã ¤!¥6çPn\:Fæ“WΧuýpK ‚JJ™Í•Q¾reŽ_5\3µ8÷P&+Ò½eç«\Ù’ÔQ7> ìÆÓ"m .Mº”k¹=µÊU§~ ;Cäd RŸ·Ç~Èfº·ŽÒÊØ⤼¬Ãå2 ×·­Ü­#Aà ¨‰y|K–B™hËå"ŽÔcÐYż}ß¾ ‹ x,h.àFÞ·ò\„^ËÅÐE*šøUr¹9±c¶ »ï%ÈIW}@dæqÛçãÌóˆk-ÃÖA6w?ŒHëR;Qj©(Ѭm@ýmOa, |NGòË[´8Ž·Ô ªŠäù7X!~PxÓGh²R…=ÐÌFh<¡õŒÏ 4ð( 6K“³ ©™Ç{˜§F§Ó€^ÊîâÓùØÆ#—40J™ìÑP‚*«8¬.â d\‰,æ×Á¸Â¦à ²qE¤"ÀlãÊó»Èž3ºñÑüg®´5V>ŸSæP-`=6χ‹x–V”t ĬZî.Š${.œ"Ìý UPvÚ ËUo ÉÒu.[›WÌ9Q l;¥.‡¹Og$ÜÝ™¢¨œEZ@K,ïÇÆ2µæCCd$•3I¨‘eê>±ŽÑ Ô;¦E.õÆ3Œç¡÷µåí[[þ?;j·up—¼ñ ˆ†¢q`+26±Ì²†¥Ž?%$‡—€eÍ0òT“Í´XJÓ}Vã‘xre‚ól€æda§ScBg’ï€ØñÕ…böÿ7 -èëYQhLDô±ˆˆ8"Šf~"š”,ËhoB{!µW-t`YJ¸v¶*®¼ƒroµgV”ÎÆ%5̪pœªjeÈ‘!²,w&»&½&ñA39ržÇEh¦wO±œ.“"Óí+㊆±¯ð>xwiÚm'¾;ÏßüLt)Ž„Õ砚Í%Íåcx4+“ò¾òžsó*Õt¾¬9~—赟:½hdz`ƒ@·Ó Žä¢v/—“¼8£öøî©é¡S»7Ì©µ³9¯3Âáp~⤤KÒxÑï€cûÎc§?Œø$uá3¥b¿?-À 4þ+…ûv°¤Jq‡´ëÒHö³µdG,®Uûˆ?Xa?N‚0bˆ»ò!Ë#Ôz~ûNšƒÃˆü é€WdEõX"q/w€þJÌ'†¬Ù¯(›ðûÛ´ ¡¤ØË\®Ø¤T‰¦d[¦ÉJ·.‹ÝþPË´Aù(O7Iݧêٶبæ]Rn²RÖdsö&ÝfwE¡_ª‹Þ^.«Zµ‡zYkòËåæ°KóZòá€x²'Û³µLgòÊš„F™ÜU¦û¢¬uâW­3¿öIYg*“Ž‹îevÐë§oocìù:#¬Ö©hMbY7×O¿’éô³òçMÆZ–§¢«ÔoàŽ}Œ¸‚ûŸM†›•‹yBv›å¬ÅæDË!€HþÛÅ^uÓÙt¢­Ú%Û­ª¾Nêå]ªG„%=“û"Ó£íÒ]Q>¨ú¾,^oÓî¼.JÕ[ «;lÅN¤z¨´ªGR«´®&§ÑÑõÒè^I£“='§ÑéåÒiU÷3«‘´8$à@¼Éž‡Db ÈÑ;yuI›Ûö8 v#ét-Ÿ2×NÈrf³;ch•òô3å$÷€â)]‹Xc„íQ!<·\'ÊÇs0ý=U€ (ß«Ï”,0þ:É Z„Dð=‘ÓE)½.]žëú"š¶ÒXíUSªyÚ— Nëí8²¥QЪk—Š9ARwiw–A$·øÆ,â(¢Á6ëdÖÇ6.³C°4飼¤ EÕÑ •Goª´TqGQAX_}‘BôæÞº #5‘ŸÍe¹u%Æñ¤¶á%©³ž$vÒ>ÐÍÎÈæÞYÞÑIGY¶,ù]ÐQ&.#XF˜9ÓV–ñªXFw=ó³ü\¤eéÂ’«ß ÜS…«ô[|_YýÒúduIòÏ ]àC&ÿ¸aéâ(|Çbç¶.}¤â ®F¦ã|–î<¼¹*nÓÔùÛüÌ!÷Jì~¤½è-™µV‹XW®¹ìB"T,†³i9ÀûûÞ×®ÔÙçn &áˆAÊ ŒÜmì_¬Möiçn‚…¶GÖ5JƆõà„Ðßÿ`È’ˆ.‚M4VûŒLà%t/é|ÚäQ=ÿ¸tOÙü IZcé9ß­vñæ Æ¼3jÌY–][©‡#ÃN¹lÅ*e×{wz¨3ú'í.5ûv~»š¨¯•Ck VÒ ¿WN…&~vŠ!X¥ÓzFH|Ga}T|o깕¦•X¢üñ”­©ê3é¥ÉDP}f:¬¡%9q¡¥’¬7Hm2U\¥àHlÊG®®.qycí^R4Ƴ.>ä™#‰¦™råêV‚ܺå¹v%‚ìùs÷¯]ͼ»Dñ¶&äíÊŒ¥¯„¸â0ÊÝür”¨Ùýú:ûÁh¶ÞÉëH¾1VʳQ‡këÊÜö_rúßnœ€”~ÞÃv*Oç¤ß·—i°±ÜÓ‹û|ââ2ÍŸî,ººì¦ef¤Å‡‘KÓÔóIƒT:§{qG“çÏ ¶Ÿ@»-Œ ›VÁ^‰–XãÈž>¼ø1ìû7—fèàÈçý`›K®Ì¦®®zWcÍH®ê†€?o‚¿ÍΙU•¯â ^-y«Ë ‘¡² Ž Ã9™Üï³Ó.§øÜ_8dO‹–Ãuìâ§*ƒQB'×™5YÇnåæèp'Ü•Ð7qFŽRD–OL©RéÈ-äWØMë^¢¨±xäÖÍ-+ÓÖØCœÞá˜ô‚åÇÉ:Ò_$q |Ž—´—ÿçZºœjõ˜¢C[Ô¥Y¾‡pú½)!n«hcÊ]iÀ¡I¢í_VÏF/«¿˜¢Ñ‹ê¯È$^5: ÑÜ4q•„½·Â¿moa=ê-›Á8ð¥>ŒcŸs>fgúâÅ“ÿ®+Yv endstream endobj 370 0 obj << /Type /Page /Contents 371 0 R /Resources 369 0 R /MediaBox [0 0 612 792] /Parent 151 0 R /Annots [ 363 0 R 364 0 R 365 0 R 366 0 R 367 0 R 368 0 R ] >> endobj 363 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.935 600.29 51.396 608.729] /A << /S /GoTo /D (noweb.NW2i2LRH-2o3xLS-1) >> >> endobj 364 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [120.695 597.652 132.156 610.717] /A << /S /GoTo /D (noweb.NW2i2LRH-2o3xLS-1) >> >> endobj 365 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [417.204 597.652 428.665 610.717] /A << /S /GoTo /D (noweb.NW2i2LRH-4XnUPC-1) >> >> endobj 366 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 201.892 51.396 210.69] /A << /S /GoTo /D (noweb.NW2i2LRH-OTtxn-1) >> >> endobj 367 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [193.436 199.255 205.399 212.32] /A << /S /GoTo /D (noweb.NW2i2LRH-OTtxn-1) >> >> endobj 368 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [417.204 199.255 428.665 212.32] /A << /S /GoTo /D (noweb.NW2i2LRH-4XnUPC-1) >> >> endobj 372 0 obj << /D [370 0 R /XYZ 125.795 791.866 null] >> endobj 373 0 obj << /D [370 0 R /XYZ 83.504 755.001 null] >> endobj 374 0 obj << /D [370 0 R /XYZ 83.504 743.045 null] >> endobj 375 0 obj << /D [370 0 R /XYZ 83.504 731.09 null] >> endobj 376 0 obj << /D [370 0 R /XYZ 83.504 719.135 null] >> endobj 377 0 obj << /D [370 0 R /XYZ 83.504 707.18 null] >> endobj 378 0 obj << /D [370 0 R /XYZ 83.504 695.225 null] >> endobj 276 0 obj << /D [370 0 R /XYZ 81.963 601.376 null] >> endobj 379 0 obj << /D [370 0 R /XYZ 72 598.648 null] >> endobj 380 0 obj << /D [370 0 R /XYZ 83.504 601.376 null] >> endobj 381 0 obj << /D [370 0 R /XYZ 83.504 589.421 null] >> endobj 382 0 obj << /D [370 0 R /XYZ 83.504 577.465 null] >> endobj 383 0 obj << /D [370 0 R /XYZ 83.504 565.51 null] >> endobj 384 0 obj << /D [370 0 R /XYZ 83.504 553.555 null] >> endobj 385 0 obj << /D [370 0 R /XYZ 83.504 541.6 null] >> endobj 386 0 obj << /D [370 0 R /XYZ 83.504 529.645 null] >> endobj 387 0 obj << /D [370 0 R /XYZ 83.504 517.69 null] >> endobj 388 0 obj << /D [370 0 R /XYZ 83.504 505.734 null] >> endobj 389 0 obj << /D [370 0 R /XYZ 83.504 493.779 null] >> endobj 390 0 obj << /D [370 0 R /XYZ 83.504 481.824 null] >> endobj 391 0 obj << /D [370 0 R /XYZ 83.504 469.869 null] >> endobj 392 0 obj << /D [370 0 R /XYZ 83.504 457.914 null] >> endobj 393 0 obj << /D [370 0 R /XYZ 83.504 445.959 null] >> endobj 394 0 obj << /D [370 0 R /XYZ 83.504 434.003 null] >> endobj 395 0 obj << /D [370 0 R /XYZ 83.504 422.048 null] >> endobj 396 0 obj << /D [370 0 R /XYZ 83.504 410.093 null] >> endobj 397 0 obj << /D [370 0 R /XYZ 83.504 398.138 null] >> endobj 398 0 obj << /D [370 0 R /XYZ 83.504 386.183 null] >> endobj 399 0 obj << /D [370 0 R /XYZ 83.504 374.228 null] >> endobj 400 0 obj << /D [370 0 R /XYZ 83.504 362.272 null] >> endobj 401 0 obj << /D [370 0 R /XYZ 83.504 350.317 null] >> endobj 402 0 obj << /D [370 0 R /XYZ 83.504 338.362 null] >> endobj 403 0 obj << /D [370 0 R /XYZ 83.504 326.407 null] >> endobj 404 0 obj << /D [370 0 R /XYZ 83.504 314.452 null] >> endobj 405 0 obj << /D [370 0 R /XYZ 83.504 302.497 null] >> endobj 277 0 obj << /D [370 0 R /XYZ 81.963 202.978 null] >> endobj 406 0 obj << /D [370 0 R /XYZ 72 200.251 null] >> endobj 407 0 obj << /D [370 0 R /XYZ 83.504 202.978 null] >> endobj 408 0 obj << /D [370 0 R /XYZ 83.504 191.023 null] >> endobj 409 0 obj << /D [370 0 R /XYZ 83.504 179.068 null] >> endobj 410 0 obj << /D [370 0 R /XYZ 83.504 167.113 null] >> endobj 411 0 obj << /D [370 0 R /XYZ 83.504 155.158 null] >> endobj 412 0 obj << /D [370 0 R /XYZ 83.504 143.202 null] >> endobj 413 0 obj << /D [370 0 R /XYZ 83.504 131.247 null] >> endobj 414 0 obj << /D [370 0 R /XYZ 83.504 119.292 null] >> endobj 415 0 obj << /D [370 0 R /XYZ 83.504 107.337 null] >> endobj 416 0 obj << /D [370 0 R /XYZ 83.504 95.382 null] >> endobj 417 0 obj << /D [370 0 R /XYZ 83.504 83.427 null] >> endobj 418 0 obj << /D [370 0 R /XYZ 83.504 71.471 null] >> endobj 369 0 obj << /Font << /F57 146 0 R /F55 149 0 R /F18 190 0 R /F73 223 0 R /F61 150 0 R >> /ProcSet [ /PDF /Text ] >> endobj 425 0 obj << /Length 2722 /Filter /FlateDecode >> stream xÚí[[oã¸}Ÿ_á—"Îl¬ïäN·-ºE;󴶳óà8JbÄ—Àv²“)úßKR%˶b3ÓE§@3)Þ¾Ëù.Ê·ƒ|ð·7yø%¶t÷ƒ²öþÍŸ/ß|ÿ“P“Iåàòf UÆ9(βœ‹Áåõàã°óO—ÿ¨^dGˆh‘™`î•è»8q5|°¥¨¯P^¡£}ƒúåÝùH 2üÑþØ'Ķ±anKÞ(ß…db‘ÁÊ­ÃW®÷×ýtðqDãfø‡ÆŽ|•´/r;ÃG”s”W( ”«$Ó5\ž˜Þ¤y^ÛÕQœ?Åù»zŠWKŽËn½¶œÚR†#Ú \§™†`šÜ23©NÙà)VM±jWO²^ªà8qŽ£à xf)>ž ‹+[,9 6"$3¢äI~€'¹Ò1O–w?))@Ú{r\Ç´ã1&Oç±üÛæ1ƒ–ÃÊÊ¥o¿Fùœd2•ë…ë2¨Úר¯£–ë4t-ƒô(P®5%y5ýšBJDBJ¶„TOžtÖäIqˆ'-yòrl™ö踑Yn H»@µä 5ü9R \¡\B¦4×o;[ù5–±s¥n‹##†8U‚Òvð•wç _¿¤5–DË%á§Z¬-Ýb%†Ø7è\?èöñ³Ørk–ÆËüûÏ lúà…՞ΧÈ/¥E"ñ%rP° ‚`ÖâPžno&Ò6óH;΢fK›Mj–­µåƒ0·«üÜIßF–ôý¨óå4: ´Û¢T)€ìQNPÚÓ1jøG{±$<]Ÿr½ZQšN;=D~‰rIûu‡D9ZÎS\¬Æjˆ¬²î¢‰µÆ"4Ž\WZ öJª(ùë©cyÔµŽ¤eš£PÌ)’DG±ŽTÞÖ^¤:Šy™æ‘ú[¡|>JHXo µOåQ•ÂJ‘0ÅÉ]{™èY*)©»1ªæÃïºÖû5¥¦A1LQ.¶DŽ<0¯4n^éq(¿3$Ãáylï©Œ˜Y…ííéP§eƒÜ‚8ðʘ`ûïNc$•“¯‡¶Ä^´•Ê$*-/ê#äz¶ LÐdÝbÅšl#$/Ùöâ÷?Iõ`™V•ªïí ©²X×ÓÉfº\¬ý »mÙ—ÙAJfÌTܨ|ÆÄ<3ZÔìê–ØK0šiR‚a¶hIMû¨SpÒ5¸ œëßÜ$¸FÜVæ3-£Çáb‘ˆ…f©#ð»[­_y2kjòcå—‘Áä)±'â ®h™JËzȸ€ê¾ŽÔøçT$F4³Š‚YZD#ÿ{†þ«é¯Q½íê­®zs”tÕ-Ë…æ/·\j˼ ÁñâuKŠ’—¯'7MØ´E©;œ‚Ñ7ÆV¨ß–h‚±% áÎвs˜þ߯°¹KÚc‰m5 ³œUç)Š¬ªzA¸E»ç€Ç`BoÙ’¤–ÿ:·"^0HÝm͘Àض¤ñƒ4Í1îqûALjhúäþA“È`\6웑Rß®w¸Ô ý¬ã€ÇêÒ£µÛ€Ù’°» @ð1‚€–D.ÑšZ4vPÒÉ<’Ž‹Èùr–óôÚ¤{örî®}óÞžE¤R$ñERþ-rcîN’Bg*ßhí7¾øòÊ¥âuTMûk!yfrCæδ3Æ)³F¹ö¯ûû›ÅS±zöÕé¦XAàÿ]ÞøßÍ]°ÆW˧P-—¾ú[hš,盢5&¶2Ðp³\…·Íf¡o±Þ„ÎËéb³Þ1ûÕx3¹Ë:7Þ²]ì¥fB;çƒÉUm&:™ÐqËxŠºðŒröRëI³Ìô¶õÔxol>ýš{½ßzå&Ó¬²îœ9#¯`“ô¥CÆI¦ÚåÞÈC»aî#7ÿºwäá&áOUdÓyuwE¤7Ÿs@mã)îå¹PZþSÄ«L·e¢œAfLÀûK0úSÍèÆZÞ+\Ù„#@.U¸jåúUSy›Ï.íËõŠÃ«‘ª^îUÏ{ÚÓY˜’§ü¿í;ÚŠ ³öé šÊUDx*û”ÁÊ`°¬m9ëm9ÙÁF Gýz*º’ -3ZQå™o{aè¦Ñªä™÷%±JäñìÂ÷±‹Øb±Å.ù‘ì’;v¬füb{è]Œ‚}Ô\F;¸ÌH¹ÌYéŸåuå4 µÛJuá a +ØÂ`4³~1g%å¤7%û(„äи~á¨2€3FŸEºé Úô¤mKª Òf½I{^Ý0¼ò/¦lv˜²Ë4›"r(,ÁvÐ7uôÍL›¾% «-ªx!|Zž œú\ºÜ;•Bƒ\+*emr=òÈy|³'E¢ ërÜû£¿žè®4"öÑ{ìn •w†ço÷_¢òw^™[·g/ͯ)ØUeüö!ŠëŽÏ£K.ã´_Î#'«©+'§%‹ÏÓ†K­tšÕŒ9©LIJå)°gsØ_< ,4Ž¼_ë$Zí¥I, º|MÁaú!Jd[5Âú'Ü«Ì©Z–>¤y‘ Ç(zžRªeiU/*O…6ô¬¡+’P;lp«úÿ”*VñÙÉÒWþR&'w™7STîmé [N•XñÐÒËå­Ì#]¼‰DÄ*aJ…hı:TòFn5CšS½ò/4X>=@—;m‹ ÿÙì˜Þof½42‚NÁAÁ?#÷!iFkWÁ/Ýç ºÖJ¥Ðcg^‰ÆAïÑ¥íÊ=à!¿¢ˆüœ>ûbÜáít'!ëòèÝk.âIð+Ré³hÒEäÃõîÖ»­IIä°¯ic"êp—-ßâiemg4ôr ùónh+uà‚8QMÔTç'öBMv+m‡‡=çÑœËxÉ>†4Ã-…X§Wþ‚Y³½B‹Bà ¶ f•§'u:bi±Pé4oc! “×ÂBŠF^fµ¥!UdÕªàÅ>{OD ªÓ/ÆH†™¾.û#Oð\þc¡*8é}›TéFÿXÈoy±sŒj[ɤ¯Àì^šàø°‡Lio!ýEx°Œ§ÜG¬qUÏMÔ¾Dûâ(£‰µ‚×æë«!ÛôÝ!w 멆$t3‡ÎÎ2b¦vÆÂxÛêœ7\~¶‘4ý€g§»¢Vòëè«×¶ÝúJG>ê¤úJ¾6Ã;¿³ŒBúér¿äŠy”âòywz¯¨Ô_›tM®XÏ.€Ïò´ ×”‘é6V;rºŸB{¢LŠE+{¢ù-o ËO]’¨)aqbغ㳑£Põú¼Ó› Yo³¶èÝ3Kõ…Ñ¢12¤ºÔ_ÉpAO ¿0²öâ>ɧ£èœ:>ŽKš2˜‘;§ÓæÎÙúŽ/ ŽÎk9È$KkZ%|„ôª¿‰£{åiç§r½š‘àSýI‡TɆ÷ç iܵ*gˆÓØò~_n%«vì$‚@Û»aØög‡{ÚŠ[Ø9^àÞð×´ï3íßIrØþU¿UróñÆfNIç¢ÛTÕ3×AŒuÑ°âé‰Q~dµ]Ik”æãj0bLd–—»<€½|ó°h2P endstream endobj 424 0 obj << /Type /Page /Contents 425 0 R /Resources 423 0 R /MediaBox [0 0 612 792] /Parent 458 0 R /Annots [ 419 0 R 420 0 R 421 0 R 422 0 R ] >> endobj 419 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [174.281 679.783 181.259 691.887] /A << /S /GoTo /D (noweb.NW2i2LRH-1ep4xx-1) >> >> endobj 420 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [44.418 555.473 51.396 563.912] /A << /S /GoTo /D (noweb.NW2i2LRH-1ep4xx-1) >> >> endobj 421 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [158.898 552.835 165.876 565.9] /A << /S /GoTo /D (noweb.NW2i2LRH-1ep4xx-1) >> >> endobj 422 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 552.835 428.665 565.9] /A << /S /GoTo /D (noweb.NW2i2LRH-OTtxn-1) >> >> endobj 426 0 obj << /D [424 0 R /XYZ 71 791.866 null] >> endobj 427 0 obj << /D [424 0 R /XYZ 83.504 755.001 null] >> endobj 428 0 obj << /D [424 0 R /XYZ 83.504 743.045 null] >> endobj 429 0 obj << /D [424 0 R /XYZ 83.504 731.09 null] >> endobj 430 0 obj << /D [424 0 R /XYZ 83.504 719.135 null] >> endobj 431 0 obj << /D [424 0 R /XYZ 83.504 707.18 null] >> endobj 432 0 obj << /D [424 0 R /XYZ 83.504 695.225 null] >> endobj 433 0 obj << /D [424 0 R /XYZ 83.504 683.27 null] >> endobj 434 0 obj << /D [424 0 R /XYZ 83.504 671.314 null] >> endobj 435 0 obj << /D [424 0 R /XYZ 83.504 659.359 null] >> endobj 436 0 obj << /D [424 0 R /XYZ 83.504 647.404 null] >> endobj 437 0 obj << /D [424 0 R /XYZ 83.504 635.449 null] >> endobj 438 0 obj << /D [424 0 R /XYZ 83.504 623.494 null] >> endobj 439 0 obj << /D [424 0 R /XYZ 83.504 611.539 null] >> endobj 440 0 obj << /D [424 0 R /XYZ 81.963 556.559 null] >> endobj 441 0 obj << /D [424 0 R /XYZ 72 553.831 null] >> endobj 442 0 obj << /D [424 0 R /XYZ 83.504 556.559 null] >> endobj 443 0 obj << /D [424 0 R /XYZ 83.504 544.603 null] >> endobj 444 0 obj << /D [424 0 R /XYZ 83.504 532.648 null] >> endobj 445 0 obj << /D [424 0 R /XYZ 83.504 520.693 null] >> endobj 446 0 obj << /D [424 0 R /XYZ 83.504 508.738 null] >> endobj 447 0 obj << /D [424 0 R /XYZ 83.504 496.783 null] >> endobj 448 0 obj << /D [424 0 R /XYZ 83.504 484.828 null] >> endobj 449 0 obj << /D [424 0 R /XYZ 83.504 472.872 null] >> endobj 450 0 obj << /D [424 0 R /XYZ 83.504 460.917 null] >> endobj 451 0 obj << /D [424 0 R /XYZ 83.504 448.962 null] >> endobj 452 0 obj << /D [424 0 R /XYZ 83.504 437.007 null] >> endobj 453 0 obj << /D [424 0 R /XYZ 83.504 425.052 null] >> endobj 454 0 obj << /D [424 0 R /XYZ 83.504 413.097 null] >> endobj 455 0 obj << /D [424 0 R /XYZ 83.504 401.141 null] >> endobj 456 0 obj << /D [424 0 R /XYZ 83.504 389.186 null] >> endobj 457 0 obj << /D [424 0 R /XYZ 83.504 377.231 null] >> endobj 423 0 obj << /Font << /F57 146 0 R /F55 149 0 R /F73 223 0 R /F61 150 0 R >> /ProcSet [ /PDF /Text ] >> endobj 465 0 obj << /Length 3862 /Filter /FlateDecode >> stream xÚík“Û¶ñ{~…¾tÊK} I¤“É4iœIì4æšÌ8ÉZ¢tëŠ'ûúë .@j©Ó!ÝR'ÙÏÀÄ.Àž±`ƒÉ€ ¾ý„¹ÿ¿ºúä³çJxÊ(ƒ«ñ ƒ$’!‹Ôàj4ø5—œ1|7çe>æ—"Iƒòêz1Z]ü~õ½é!pj¦yÝ\JJ–ÚЋR‹`æ^4?d•în™¯–‹ùªx=Íí_Æ‹Òþe¸˜-oªb>±?«k÷ÀoL±l¹,ïŠYVåæ'·O,«*/ ó>àÆSŒ›L’PèØà¸ÕÏî‚¡ŠtóÔ³‹ËHð´À /ë·ÍxµS3c—\‹ðBKŠ`žO²ªX×]HL{{–•“bžMÝíâM>-®‹ ŒváÏSƒ¿h›O_íÄž…J¦ÍCÙ|dAÕÊBYfeUX"™¹Zf«E ÞNØè>qÈÕ<õöÂÌRX^p΃° /.KÝôð׷˼48d³¼™>•†ÜLLŸ£;ó w"TNñ’ék:<5o§É é4¸Dj¨nr3Ëç0BÕÂýoÿ®ƒñÍ|X‹¹ýeiÔ\ ¼ïEN%a’jì8¢.Õë"›[ôZð;$i˜$í§…w|È·0\¬³²Èê…“™Énæ¼!Ù‡@GQÈY‹µéíý_/œXo¨Úb$ÙƒP83ô%MæÕäaÌt‡Â#iVh™s`5˜QVe»@¤:q áÝÎaþ%î×Ç¡HeóÐí®~TË‘aÊdðS¾ÎËzõÕ8æX eŽõšðìò¨Ù³é°ó©ŠÙrš×Tœ,G†Y¨/FùjX¯›ûÕu±²W«¦=¬{mE@ó? ž†:Ž£óK³¢U˜:T"±0S;ÆDn-ýúÁ†äí#1ïŽù¤Qó„²ᬳ àX‡L·¾bê´ý8ÅoLpQM›õ¿7ƒæê…•6‹FeØ¿Œ¤uÃiøÖ|„jòAYÿ´—ÿÙQ¨ƒL’¶dþþëà×ËXñà;óá“`íØ´iÃu mwì_‡îàÄÐ]½P@TÐ^C»€víʘé.Áç$hE:ø!aQ\B{ƒ¥™ƒXÖÝI ãfDFVKñœ·(Èà×CÔЦðáS˜€Ô0ªº- Á3öùÊŠ/PIˆvÙâ—+h+Dª–l hîÉÇŠêÕa>ÖZšN£zþI¦…ß³–V›ÉYÁ¢ýk½ÖIV0s £D fˆÖôjÏÜ>ŒÕqÄc'ÂXYðÃSò7/fjyó Zþm¹{ê–ÉÚ[Àñ††‹L[!ÒÜß‹„vwFT£/`2爧ˆg¼yö.Qr(!JZB<èé×Ã{眧L9e¿á®¿öâkCÇî{ÇHDŒüàK€üˆ>”F®nT¯® Ë‘b6ô–·^"Ô_Cˆ­¤€‡€e%ˆ˜Gl5«‰òØÉ,K ¯Û•L¥uoa´´×T]‹dì  Sæ8 ãŽ×Z:zƒ(ß¿FºÐ›ïÞ£·í͆bËEÔy°¡¨#JÆÐ>‡ö;h¯ ýú H'Ið±žÚ ´D‹AeZ®Ö€™RÐΖ°œl‚¬ K¥›åÑÐêÂqDðI‹=EvSú Ý¤´¯Ýä¥| Èu &q¨°U¾H-x#SŽ¥Y ·„¶€ö‰þŸ㈑0O0‰æÙ²åÄÝ¡éïÇN˜Õ]ã²5 9àÒÜBë©!s–õ{T " ÇtôÖEÑ´%´›§Ñ‚³Ž í\Ñ>©·˜Ë¡d™œYZæY‚ÂUkÿÝ?üˆË 5üp‰xÿ}Ìéå@R¶Ë W—<Ù‰„މġô’ž/½ü ×_‰^~Fc[Œé4²y~J%ؾBÑ­ÓˆêvEB ú <q(Åœ€‡íç¨4)>1‡v íKÐ9~„ëûÇB*ÍÀ”À½$ˆp *—LüŦ.=ÅfZMò‚³¾RíÛ S'um¥JMåÏPHNZßf½‘²@*+™6¼¼Ãå_#æð¡2E\Þ²”[ª9È`JçÐÒø7;ZMt\¯àºò%=îK‰~‚µõr fÈÏŸSú­hu‰óm Oè]K,u,Îoò@fÎcÁ9-{<˜M‹§gÓ.°xE­Œ#ä÷MZóx¿#«–ÒzÄ^•!4"û¤ åJMç­Ì¶Œž=úŽÑƒµ$ý€Ñó‡ãûƒöŠÊäƒHEgAöð…”3õ(´6Eñé·HÚÍÌ—}NÃ>䉰±è<´Íæ‹Þ´~ठ;m„Ñ‚0®Xð«§Þf™€p!™•ók’ aé“ÕÔk–Ç+ª%<º‡;7Î)³íÆï2¯‡RµBOåQúzR)ÕLÁÎ…{Zû×zoþäžrOË=·-/õv¢eãø[P&¸¤ÜGéÞ|ÔÜ“Ÿ÷L•—Óä±Y"qÊh­®”Èê’ŒhõIˆ\›ÞRu÷O³2E)ŸÊÅîOù<³RiÇ&K”£Žb«ªÌgœ³²Döwµ#Dv¨ÓLˆsá ·9_ØÃí/œ‹fí‘ÎÌ‘”oø€lÓ­³^ø iƒeIf¥ƒÓ2‰È…Ó Hýû¡†ëÿÍ”¤ûf2탉vs&w’h|mâDbÖ":«Õ†$§ ¿úÓjmë+à”Ú_"ñŸ»Ô¢AkmvSIIµ§JÅ=˜ËfÆI³b$šY»ï+A»„š­kÊR‘sÜX·÷„j3´¥sÃó’ÎfåÌ9È·t&ÈÓvî_#_î.OÃÁ OG’;0¯äÑ›2m¾âéüäž™%4*¦ T1Ó{T̘TŤ6-S—´Œµ¢ª50›ûd¼a„øŽÛ¬ï”æé&ØÛͶ$<‘Ÿ»Vv$ûñpì/]ïH'>˜qÆgÀ8ÓDÄNG´EŒŽ(&ê¥à[ù ‡¼âÁ[o‹.ôö—ä}Jß"JxõéÁÇ­=ØDŠ†´›î·ÒC¶=lÂH†îl¥‡´©ñÍ[$K?9©ÜaŸBуÉH1Q2’ÑjJôqññ܃6×sD ýÅé ÌÃk¦wv¥Ü­ˆ·Vë'¬ØÈ€íš8¦ÙÉQRVèH:[•+”{ƒkFàë Mxà7…VIÇ[²Dé·Ù1ªcÈèCpÇd—¨&KébMÖ®"µŸJ*¯ûÑ |ú©3ç 3Þgäz;ü!×(ÞIöóe;dÁOˆÛMО)¨€õÔnÛÞa"W_ŸRçx˜ÎŸÂ®//h´]}†®8ƒÍ£Üɺ˜Ÿ-¼²pwˆ²†È^?PY%AéTiÂÔ{ „=BÁλQiÄ3ªŽzppq'à—Hä—Hè¼£¢Ÿ'ªµ·)¦WÂœ„Iv:±5S ÿúåHøÕØÖ€Up܇,úÓ€ã*“Ȉ,ù¶5à®Ö{¿áýÈHM·êþ¤“¤½¹?¥ÊÏé%êhKoo¡Ý¹¿‡&/¼XÜ=ÉQK~.áš²3Ôƒµkõƒþ;òj]ÇV-\z%Þ—|x:±"ÎÎâÊkïm!{í!)På´>@æ4ø›÷+OÁÇÄYh³æƒ¾ÊÖTß±üåoÉDEW×å%UG_PÅ蛸ü& ¼×·òÏÑ_­ô‹÷óŸz«/¼Ùí§{a«¼±ÞÂA¹­‘·ÞzÜÞÂMøšs±Ø7:A$O£ŒŒN¤ŒPüjꯠ½Aµö§¨½õ¯¸ÿì„Ý‘ªSù„§alŸ&”%ÈO\IÔž{$ÝÎýí=Í>òÞö@h§´ŒÚ34ë¦"¬jØ5½q¬´ScÈï1£a‘ê4X“Q#þµÏvÖÃÎø/¨-¶=¤¼tÅÛ§PG{ ××®²¶w5m2Õðgê9ÆtOÑw ¢Ww€¿~wˆ­îïsL›Î(†|h¾½¡ØMÈ_öòw~¾~ÒªÜV´S-~wŒŽ;“Õú7êªÙ$õÚJ¾ÝS 'H—Úw+œ-qýwÉz"µCÛ9SÙsà< ([¹®ÛÉÔx òAøGìˆI™éÖ «U÷”ù ;³¹êœ ¿Ùò5¢òãû“œPÒò÷Œ¦­ÄýDû“öv›q’’âdI"n &©ÈH\²Á%¬`õy2ü.1q"õ”ez2º ž¤X{Ú×æD°iûñô!›£B–ÞzíÏ:œk:?;}IL÷³Y~…ÄK4›åÖÆù˜ôЯVò=H% ¾„ç¿tÇ©øªŠ3*¿ü\Ñ9'ý¾´ïÁ!qýä(€Ôó¡ÐýÊ q¢Ü€ ÈÌ¥·KÏÓb§¡FÝpìgÏÙ…Èâ lªÿìyÌÑ2L“öÉrf¾¥0ëf¸XÞ–Å亲¯4RóJÕ¯ÄiÈtÔ¼£2ûH¬£X7O[Õý:ç,ÔLsÀXêPÖCÔ¡r×QÈß\}òO´Ë/ endstream endobj 464 0 obj << /Type /Page /Contents 465 0 R /Resources 463 0 R /MediaBox [0 0 612 792] /Parent 458 0 R /Annots [ 459 0 R 460 0 R 461 0 R ] >> endobj 459 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [44.418 637.927 51.396 646.367] /A << /S /GoTo /D (noweb.NW2i2LRH-2yIl95-1) >> >> endobj 460 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [144.345 635.29 151.323 648.355] /A << /S /GoTo /D (noweb.NW2i2LRH-2yIl95-1) >> >> endobj 461 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 145.364 165.939 157.469] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 466 0 obj << /D [464 0 R /XYZ 125.795 791.866 null] >> endobj 14 0 obj << /D [464 0 R /XYZ 72 754.004 null] >> endobj 355 0 obj << /D [464 0 R /XYZ 81.963 639.013 null] >> endobj 467 0 obj << /D [464 0 R /XYZ 72 636.286 null] >> endobj 468 0 obj << /D [464 0 R /XYZ 83.504 639.013 null] >> endobj 469 0 obj << /D [464 0 R /XYZ 83.504 627.058 null] >> endobj 470 0 obj << /D [464 0 R /XYZ 83.504 615.103 null] >> endobj 471 0 obj << /D [464 0 R /XYZ 83.504 603.148 null] >> endobj 472 0 obj << /D [464 0 R /XYZ 83.504 591.193 null] >> endobj 473 0 obj << /D [464 0 R /XYZ 83.504 579.237 null] >> endobj 474 0 obj << /D [464 0 R /XYZ 83.504 567.282 null] >> endobj 475 0 obj << /D [464 0 R /XYZ 83.504 555.327 null] >> endobj 476 0 obj << /D [464 0 R /XYZ 83.504 543.372 null] >> endobj 477 0 obj << /D [464 0 R /XYZ 83.504 531.417 null] >> endobj 478 0 obj << /D [464 0 R /XYZ 83.504 519.462 null] >> endobj 479 0 obj << /D [464 0 R /XYZ 83.504 507.506 null] >> endobj 480 0 obj << /D [464 0 R /XYZ 83.504 495.551 null] >> endobj 481 0 obj << /D [464 0 R /XYZ 83.504 483.596 null] >> endobj 482 0 obj << /D [464 0 R /XYZ 83.504 471.641 null] >> endobj 483 0 obj << /D [464 0 R /XYZ 83.504 459.686 null] >> endobj 484 0 obj << /D [464 0 R /XYZ 83.504 447.731 null] >> endobj 485 0 obj << /D [464 0 R /XYZ 83.504 435.775 null] >> endobj 486 0 obj << /D [464 0 R /XYZ 83.504 423.82 null] >> endobj 487 0 obj << /D [464 0 R /XYZ 83.504 411.865 null] >> endobj 488 0 obj << /D [464 0 R /XYZ 83.504 399.91 null] >> endobj 489 0 obj << /D [464 0 R /XYZ 83.504 387.955 null] >> endobj 490 0 obj << /D [464 0 R /XYZ 83.504 376 null] >> endobj 491 0 obj << /D [464 0 R /XYZ 83.504 364.044 null] >> endobj 492 0 obj << /D [464 0 R /XYZ 83.504 352.089 null] >> endobj 493 0 obj << /D [464 0 R /XYZ 83.504 340.134 null] >> endobj 494 0 obj << /D [464 0 R /XYZ 83.504 328.179 null] >> endobj 495 0 obj << /D [464 0 R /XYZ 83.504 316.224 null] >> endobj 496 0 obj << /D [464 0 R /XYZ 83.504 304.269 null] >> endobj 497 0 obj << /D [464 0 R /XYZ 83.504 292.313 null] >> endobj 498 0 obj << /D [464 0 R /XYZ 83.504 280.358 null] >> endobj 499 0 obj << /D [464 0 R /XYZ 83.504 268.403 null] >> endobj 500 0 obj << /D [464 0 R /XYZ 83.504 256.448 null] >> endobj 501 0 obj << /D [464 0 R /XYZ 83.504 244.493 null] >> endobj 502 0 obj << /D [464 0 R /XYZ 83.504 232.538 null] >> endobj 503 0 obj << /D [464 0 R /XYZ 83.504 220.582 null] >> endobj 504 0 obj << /D [464 0 R /XYZ 83.504 208.627 null] >> endobj 505 0 obj << /D [464 0 R /XYZ 83.504 196.672 null] >> endobj 506 0 obj << /D [464 0 R /XYZ 83.504 184.717 null] >> endobj 507 0 obj << /D [464 0 R /XYZ 83.504 172.762 null] >> endobj 508 0 obj << /D [464 0 R /XYZ 83.504 160.807 null] >> endobj 463 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F18 190 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 514 0 obj << /Length 3869 /Filter /FlateDecode >> stream xÚݯ۶ñÿ| cÀ0½$VDR¤ÄÝ€µI–5íÚõašƒbëÙjlË•í4¯ÃöÙw<’%˶ló¡Ù„–(òx¼ßwR4š¢ÑËG‘ùûÇÛGÏ^ðdD¢PF’ŒnïF %1 £˜n§£7Á×åöf̨²ÅBwŠÕ]^å«I®.óí¼œnÌ Ê\”Ëu¶-Þ.Ìí¼3Ï¢x—/ŠyYNõï»Ýj²-ÊÕæ)üfi‡³0Ô·òÙdÛ»za.W‹û›nÿ ÛòXjð×åfc  ÜBAâàe¶ƒÙJ_o@c¯Ì²šæ•înKýä:¯îÊj©/ÖÐ(˜iäÙÄ, Ñ¢G­ò—ßgUQî6úúº*×ã>¨ój[ä0ˆ&iPÞé¿Ûy®;.ÞÔooê÷&ÇŸúGªùáåˆPrNÕ!ÃrŒ‡4Mõr,$7cEQð\#Z=ùªF´ž E%0ảžàEY©‡¸ƒVõ«TÃ÷¿ÑST?«|º›è kD«¿Š„vÛb5Ó?—¹ž6²ÕÔŽPèÌêi4ªx™'v‹m#¶y–] ÁÏóBŸ &öÚÛ\ÿ–+ÓCJRÓ·ÍßM±\/Ì€e¶­ŠºŸ-fùÛ*S4ÇÁíÜ cŸUÙR㕤.^™HˆH€!4=ƒÿx*á²ï RYÙñÓ¨v Ê1Ö!+ <•€æy±Qì%EðêN_*¶æïFÿ¯æSÝ׬£z™þÓ"CømÉPÿ*\+»P¶²½\w,æ×Õ5ënv¹YÛ>“W´`ÃäݳT´ \¿ÍÑ*²$rQ& bÄ ì'3DºCD(±#þñ¨a¢fPÒ˜ÙAwzDÚZ‰…‚Õ#~ü¢o%`J–¤ÎRÀ‰œ‹à³¾ùh pQ;øk3„´è!äR@+ ³SFQâPÊd4vFön!yZ¯øcßD@€”5 ¢LE˜Äõ4OûáIyò¿õm-ñÅ:[#½<FT¶¶VYŠ €±Vö/N«9Æ”¤ˆL†œ =4³àHXÐçzˆ ]R‰O0u³v62 wM8½fÉ°Žˆx¤¨þ½ÙÞ—zc&W›€%2$BíK©³/ÒA œÌ¨št篈îÌ¡ÔÔ¤?Ž“`­VØŸ`»Å¶Àë%öW­eFoÆ‚GÁ››±ÁPFƒ ¶ð¸à)ÎÊdÁ÷ê’ ¦x…í[u# ~€–EÀ:0Š˜…5@ÄçØ~ÀëYЄi6ó$@;%¸ÇÀJ%,LœJçƒ@‚Ÿ‡>Ì os5Ã-0¼ož<·ßÕ7(Q ¢GEjAδz”æôQAH[8î›%§óë~ÈG¡9~Ýí%€¤ ¿0Áñ»ŠÌÅx@1PŒÃùX(¤0d8ŽSC9¶UÝO[$œûY8ÆÅ"¤ T~ÐÅE>&ˆ^"1ãe:0žFž,,C;÷ƒG¡ …cÉj‰“ °Ø&æx3Ãø>–L´„Ñ”“*î­éĽ>ÇV ¹ûÓËÃ`͆~ПגV(Qí[-™·Øú!èXz¢=x F5ÝvPÙ>´Kls¼¾Å¾¾[`û ^×#·Îõ¯¯NÊÀIªZK.Ƈ\d¸+àÏ;?('ˆ°Õœô%á"~¹"’à2ê×Ȩ/–~Ží-^ÿöÿ‚íØ~çÕÕªô« ä š!£š!ÑÙ¡1B:Üܦ(ò¨ŽÝ«Óâ(19³OŽ£”äJï)·YÅ&h´¯MÅw*\|jêF^Xôx€æÖÌE@Xôa× 10ih&ØÎ±Í dIÁ0×±Pk ®TØÅ°%ìhà¶Â¨¢‡¢ÈÔ‰òlsÒØo…h0«N¸˜G1§Æ4Ðáæé%§h¡M=ôÌ°ºãÈ‘ÓWyç êƒ_b»Þà iaÐ=rtc—¦R‰bn\rͦšië¾Á( ~×ZåØXôhcáx‚‰éN¤¯k\™5%àZ7ö ehšæK{”Ã%-EÙKAµû>ñ^Z!¥•¾”Wî¤YÖ&`Ð&ÎIo}ù(D«5ýÄ(8òOä\Jèe’íÌÞ^v§Gftä­uZŽé—ªnuð.îÁQ&?Êéä ©@|;’:4Ç5„˜¸yÂ&ÎbãzBJ;FÞU‹ “ðtS™÷þ¦OL lQUCokæ>w•œL]‰´—N,rw`oYOÂùšeM´2qAn´2ÙKÙtÐt°¾bûúÊ&„óeˆ ];$>ßiÜÔ«í"5W~¾!.NâÊÚ7¿A±rsë_ ÅZ×7FM:Ñð/&¼ÒµÃR[1ÀLÿ‚“êuS’á[ýràŽ´d‹<_'ØYGO:’|lÔªÍZÃCàÝý 7ø ¶ÿ¾QvãS]m1Ç¡÷8t‡€š‹"NAÝ¢ˆ¨g*VBáË(ƒ½äNÎ&s ˜seë±>dËÄÉr¸AýÊ‘-Y­¨R¯å pˆKÇtíÖ}ðã~§Ãézùt­ƒÓ6bšY œ${Ôiê¦&¤~4¾FÇ8Tmh\øK$ôaÝS~ì=Òiî8ºšŠËžÇÙôk-iy¶Ö¢Ñ ’wÔìÆQ³TG‰N’?3Êùƒ!Ì.ÉKCò0„bûøF]ì•ÕûÁž¾ØÉe¤›Hð|<ùÌZZG¤Àvc„€§¬‡+ù+§ˆ1ëø‰¿€¯× ‡÷2»%ÃeûëÁVØĸMeèâ°F•1¶Ø3$è.7ñàɉõ„á¥ûžH>bµ1÷VN ”ɦóéuq%šøïŸãáÎ÷oQ»³¶ZNƽ®‘¨‰­ qJ_¼3´‘˜ðÄ)"ºÈðJ}‰šäáJ%u]RìˆIÖ+&/µæ(=¥ÚdÇÔ¡›¶ºLln-5%ÏM½öÆqšÎyh'ûŸŽÙë:Ù tK4k”¯­×ý{ì‘e—mKì>µqê(òôYž›Úa'l¸…`Œ ŸBcórH‰|ª”wí!9v‘œèÄSrlÌq–”¤¾´¼4Hî’ã˜x²iS2ã-}+bõnÂnií.Z“–V”=­‹ý8fn$˜•]K'®¹íÔ’¥ÚI'ÊD¹Rxxôïú¥ $Üv 4·Ï kàÆÌxpOñÁVs;tžWýil VÂÑ•ç¥áZZpø3”#õî©'ⳇwuEVÄ«½*êª+Œ“õÞk''µq”}ß›Jsdôžé>ç /œjÔ'¡‘Ö`£¿Æ‹W‘4e$Nü¾ö5Ñg¾˜ì”“røaÙ¤Ì0‰)1º\À¨šáKðañ™«åÝ“…AQ-Õ‡-A‚ǃ˃ÔËbèÌR&(†ˆÏǃñÿå`xœƒ+;9?ä”ú®99Rµ£ ‚FAéJð¥ øC»ÃvO•"‚ßãJäÊ°8pžD Ö®2á«~=; {}§.s¬ ­¯`Ѽ›ÐÜÚ̉I­ZqçzQÏ–)ûzQï×dX•Ç”5MÐ74ì¨-L–ýÞÆ„Q}Õoãäb6‚§F4 †bŽbÈ—sÁ©Üa UåÃrONGµý»C¸–‡"e§Âm‡_b<‰}¼p˜cv,QûúÆÌLpÙNOmí‰Æï ‡?>U¶hÎFgHg'â9T¸‡‡Ã¸©Ñv7)x7f‘×ÍQçJ±e:•2ruù''¬æÔÄwØ.늰¾*0Ò¼Ût¨ ¦Sõh¹ihfPï·²£Y>…R’ãzO§F°$¢Æ_ ›.(šïG +°pü8&{ü81Ìä *оEQ¢8åI§°9…}ñÉ‚U==«ÇKúLÁªó/õ|¬‹«cf Ší¥âÆPÁÐþ~ç¶Ûá‰fCmzâW°àuõÜ.u[§E\kÕYêî¾iâ5¥ºõ<§(H5Ⱦ¼rØ r)Èâ[·Pàœ‚€^Ó"ù¿•i!­°1(ÎŒas©ØLýˆMy N-› <|Wx¢Q4>!ýÈ^ý1;-¦®©?n៨?Nz¿/X ‰²ƒU!Ë*ùQÈ*ûM¯ ,Ü8­€…‡<Ì«9‡þv?•Ji²ÄV¢?QÚóq…ç·þ P endstream endobj 513 0 obj << /Type /Page /Contents 514 0 R /Resources 512 0 R /MediaBox [0 0 612 792] /Parent 458 0 R /Annots [ 462 0 R 509 0 R 510 0 R 511 0 R ] >> endobj 462 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [264.989 712.813 273.047 725.114] /A << /S /GoTo /D (section.4) >> >> endobj 509 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [44.418 592.948 51.396 601.387] /A << /S /GoTo /D (noweb.NW2i2LRH-4ZZkMy-1) >> >> endobj 510 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [146.167 590.31 153.145 603.375] /A << /S /GoTo /D (noweb.NW2i2LRH-4ZZkMy-1) >> >> endobj 511 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 506.861 165.939 518.965] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 515 0 obj << /D [513 0 R /XYZ 71 791.866 null] >> endobj 18 0 obj << /D [513 0 R /XYZ 72 698.138 null] >> endobj 356 0 obj << /D [513 0 R /XYZ 81.963 594.034 null] >> endobj 516 0 obj << /D [513 0 R /XYZ 72 591.306 null] >> endobj 517 0 obj << /D [513 0 R /XYZ 83.504 594.034 null] >> endobj 518 0 obj << /D [513 0 R /XYZ 83.504 582.079 null] >> endobj 519 0 obj << /D [513 0 R /XYZ 83.504 570.123 null] >> endobj 520 0 obj << /D [513 0 R /XYZ 83.504 558.168 null] >> endobj 521 0 obj << /D [513 0 R /XYZ 83.504 546.213 null] >> endobj 522 0 obj << /D [513 0 R /XYZ 83.504 534.258 null] >> endobj 523 0 obj << /D [513 0 R /XYZ 83.504 522.303 null] >> endobj 524 0 obj << /D [513 0 R /XYZ 83.504 510.348 null] >> endobj 525 0 obj << /D [513 0 R /XYZ 83.504 498.392 null] >> endobj 526 0 obj << /D [513 0 R /XYZ 83.504 486.437 null] >> endobj 527 0 obj << /D [513 0 R /XYZ 83.504 474.482 null] >> endobj 528 0 obj << /D [513 0 R /XYZ 83.504 462.527 null] >> endobj 529 0 obj << /D [513 0 R /XYZ 83.504 450.572 null] >> endobj 530 0 obj << /D [513 0 R /XYZ 83.504 438.616 null] >> endobj 531 0 obj << /D [513 0 R /XYZ 83.504 426.661 null] >> endobj 532 0 obj << /D [513 0 R /XYZ 83.504 414.706 null] >> endobj 533 0 obj << /D [513 0 R /XYZ 83.504 402.751 null] >> endobj 534 0 obj << /D [513 0 R /XYZ 83.504 390.796 null] >> endobj 535 0 obj << /D [513 0 R /XYZ 83.504 378.841 null] >> endobj 536 0 obj << /D [513 0 R /XYZ 83.504 366.885 null] >> endobj 537 0 obj << /D [513 0 R /XYZ 83.504 354.93 null] >> endobj 538 0 obj << /D [513 0 R /XYZ 83.504 342.975 null] >> endobj 539 0 obj << /D [513 0 R /XYZ 83.504 331.02 null] >> endobj 540 0 obj << /D [513 0 R /XYZ 83.504 319.065 null] >> endobj 541 0 obj << /D [513 0 R /XYZ 83.504 307.11 null] >> endobj 542 0 obj << /D [513 0 R /XYZ 83.504 295.154 null] >> endobj 543 0 obj << /D [513 0 R /XYZ 83.504 283.199 null] >> endobj 544 0 obj << /D [513 0 R /XYZ 83.504 271.244 null] >> endobj 545 0 obj << /D [513 0 R /XYZ 83.504 259.289 null] >> endobj 546 0 obj << /D [513 0 R /XYZ 83.504 247.334 null] >> endobj 547 0 obj << /D [513 0 R /XYZ 83.504 235.379 null] >> endobj 548 0 obj << /D [513 0 R /XYZ 83.504 223.423 null] >> endobj 549 0 obj << /D [513 0 R /XYZ 83.504 211.468 null] >> endobj 550 0 obj << /D [513 0 R /XYZ 83.504 199.513 null] >> endobj 551 0 obj << /D [513 0 R /XYZ 83.504 187.558 null] >> endobj 552 0 obj << /D [513 0 R /XYZ 83.504 175.603 null] >> endobj 553 0 obj << /D [513 0 R /XYZ 83.504 163.648 null] >> endobj 554 0 obj << /D [513 0 R /XYZ 83.504 151.692 null] >> endobj 555 0 obj << /D [513 0 R /XYZ 83.504 139.737 null] >> endobj 556 0 obj << /D [513 0 R /XYZ 83.504 127.782 null] >> endobj 557 0 obj << /D [513 0 R /XYZ 83.504 115.827 null] >> endobj 512 0 obj << /Font << /F57 146 0 R /F59 148 0 R /F18 190 0 R /F70 201 0 R /F79 203 0 R /F80 204 0 R /F81 206 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 562 0 obj << /Length 3761 /Filter /FlateDecode >> stream xÚí\YsÇ~ׯÀ›Á²0žûˆK);ÙåØqÙ ËIYÖ,Ä•pPHŠÿ>=Ç3˜– HSŽ_D`·5g_Ó1‰‡Ór6+ÖÅr[ŽßÌ ÿl^¾/æåÅj5Ý<‡'š4=*n{´BãvA¬yV©íEhórµÙëÒwK†onC+þÏ·ã«Í¦»‘Û¡†7~¨Œpê':Þú¿å6Ì|±šáÕ2È@ÿ“‹rù6‘œ\­¯ÇÛ«uáׇèöúPX)4tî:-—³0g´È-§THrYI#7î51ÌaøQhÞ½ûçjÆ7Y-'ëb[ÌoíÆ=÷ݹA0}÷i5K^5kçF¢p4pÌæªʇ c¢Ñ"IH%ñ8L)âˆrV ͼ„ŽºbH²ZâÝßs=þ2¥[]‘³‘rø"×å0.Z ÿDH[D#a$ük@.4‰qn82F F-Éì8ºîñ]v=£¬™•ÙíGŠ×Í<ÏG‹zÉÉMM¥³dj$×™F˜šhjeÐè·åu” *3 PI ÿ[ FŒ ­ÒIÅ{gRõ¤^äF»KU­jã5X‘j¸(ƒËQÜÎJa¿ J#…SÑ–Äc$ÖǯkÄ~Ã,;@h»YõFÓ…ßUce:uÀ©hfª¶Þìðˆs>¼É_"&k·&K†•ÅwM„1°Óg&üŽ3qÝæ÷¶Ü,³–þ?¸y?{yþìÃ3«©Ð7X$AŠÈ)I“ųW¯ñ` /Á/#ÎäàƉ.À°ði>ø÷³ŸCüR±u‡¶8G4ÃõIóûJ¡/…¸f÷Ý[…0®Ýò(»_ÉFÍyC0&£ Jµ´Ûol«ù×\g%_Kìûà X«-˜¿1í¶Fõ6B];’×⾎W¬žËH.#K´“²º¹ÈWìÑÖ(T xPï¶;ƒ€Á 83pCAî¿™Ö0DaXup@š…¥(3˽±]w¯8MV< Bp;lfB?óeÇbR¯çe>Ds¦îäÛn33‡V4À¤ç6Caµ©Î@¸£%<2ë}2½¥j™ô¹Ü¦9ƒ$ÈIDL“\k#44jɽð^o”Û!P,Ö,Y½;‰žbS;ðW¹¥€ÕŸÎ퓧qè$ãyº›9Ñhnszâ´ ßUEöOèun—TV ˆ „q7nZ­˜>„ Fù Ü‹žûìB¥Ð>RS8DIëeøo_Õcð Þ”@‚AüCÚ°8RñÈch‹hZRU„Ã(GÁñKø•î‹¢•ƒæ­0uÙ4š\!¸¯f³:$5S aø›‚Y“Rˆ.¥1LUäüŸ³Ô¢Ú”®°¥aM̨E÷F-5vWëf>ÏŽfOêÑü' p±ZëoÊíEVe,§JŠWƒ«’±_Ë:6X!›œ´äÀq*ò@´€Á®Ùµ0V.†)’€z($lTìQ®¦) Ù©”+30ƒ4(iRóh„ qˆ×IJ>bj&`“ƒS=Œ{ЀÉ<`F‰a lŠõµg]ÔpìþÊá庘”˹Ç3ϾȊ•Q1ïÉ"„Bb×C#‚ƒš¶‰.cÚ'ï7,üJSßË¡„7ØkÄ îÜCH°” áˆ4ï刈ä©ÿØá@¤Úâј½f3Ù @5R`–­1;Ìî“ßbø°€4½\Žçö›hQ{ùT’YVÏî׎ى©¾ÈÁtëA}4ÒUÄýW>9Òš[^£{hhEû@+ö08ýA U/£IvÏm/“ŽR%³ˆ7Ô“÷£ —q€Ø&ælkØi£ê`§”%Uªm,Õ¶Öæ3ÒUT“‹þšÃ€0ãþÃ×YåV²8¤w(8Äi«¹,Rð‘€'dÚå*¼¢¶(Ãy¢qasVG™œ†;*Õد>7¼¹(օ׉›ð÷j>—¥†tÔfµžkÿñüLƒÒÝÎWá{ññr¼ ¡ ¾ÚcÆuÓ†;ÈÙ•†D@ñCïˆKû8Që¾uʯElͶ=án³ƒ‰ –­>OEÀvqÍÞpCŸg¢ `ÆrΗD+‡ëøÐb* ÎÈqÙî/4%¥cL»yKðZYI>:oÙ]óÔ%kANæ2BC¿d—¥¹/ðò¼eÓÊÓ]E–’ü`h‚øEÔéíêDÕÁ%7Šøà29󅚀x¤†Ë¢˜†·à«ËkxåÒûàêÒËlWþûö¢\Oý£àÙíS—•ø×…YéJ{$F·Ûp”«,¬˜Ó«uýVà‹›í*{n¼(¶«iÒ#Ü ÷gµÆˆ8ÂÅ¢Ð&‚ï;R,2À!éð…áîFD,@Ò%ÈÓÄûÌAàHj9m ±÷ ò'w>Õ&J$´ÏhÍ‘ÍòÔˆ„ÿÃ,Z‹XÅ ‹„¬`!«’üÕÛ®£³·}jLfSÃÜœ„‹Jâ}-ïp ‹ŸûSXâþÊ! ßY°m Ûe¬=cQºI³ƒÃ z ´ö\ÊG2 Pa n;5ŠTá-ík\[iu@á=W¸GãCc z «¾çœã|í´Ê²68†¹ð×Úe{-›0láRž~ó<`ØU[íivÚu/³æ}Íú–D¶jaNaÖ:5kê͘a† '±±á ‡Åõx~å“ìviT]74‡wËðp¾štSv ¹á{³8’[,“+UzãhÇðÍ'›tåÍòÜæ‰áf/‘çªÎëy4¤•/¤Ï盪ÂBàMä6Ť&zȫݬkì쉰ý殎Ö9‘XZæ?ÖçƒS»1€é†Sâ>î¡ÕöÚxá_À¸íø.n/ƒÀmø³ðýÕëðá3_ovF†ru)"LÌ|ûñ“È—7Ót)§š$}í‡üÂO¦$ÔT:%YMÉÍ'¼+ûÍËa]l¯Öa/‹ãÉv~[«÷¦h)Ü5ÕlMõ$€AÆÝ!B(žd¡xòåÇKP“q£(?­W—ã· 'iëÈ‘s:|3Þ”û‘ Ëi1öméž}ub_´;qv_þä3_+`y5ûÜbeû}]¸¥ «N—«å¨æìÝ«¶mX‰m±^lr6˜ ãOèL…?æ™ tZ;"W¥JLsL2bgWKçg6ùS#Ùð]ÛŽ Ã«þê5óàÏü˃劫<èc¼_‡¦Ýag}d+-æ™þ }ìÖ:0fnâ†Ï.>f-|hÄöØ£=«&ÃÇYÜCÙ¥£eZR9^p-ø¸ŸVcº„ÿÙ6ïA«5MÁúcÓjø­Ö©êø•jgedžË°­e¥r¯*jq4;ìcL„¥¬“±žßº—ñôÝÕfž9:^Ž×ãEÎ=ït×èˆt½“&pÏ{8Œ»mÐ×.ÝÑùâ·ª¸˜ÍÕä¢ ø;Èv¶šÏW7M‰ÿÔ^`Ø€p±šO7q°YhæÌ‘dèc-a´ë¨µmȽ#À‰ ðXrB¶î8`’_.þßaÀ»‘]-0} Þqgâ)cߺ¹ÒUoo+TÔÛáj<ú8H]]ÍL7ÝS5ÖÇóħá—Ï#U}ÀõÈGv=¤ëy¢žç¨Ã›ûzž?ó¼‡+Œ‰¶ìl¿¸-HZÐV+Ëò#ED«ôuÄpà ÆÃÍjdI3WŽÏ'ãk‡–ì«i¹Ù®Ë7WÝ­e¢%I.~z怺ã"¸V&Mr1nÎßC<"Vsý'×jR¾u;épPý¾ãÆ•Ñ™ EšµE¦"Ä’ŸsGÄ![äΪŠÑw™šÀáªVs™Q!œØÀaÍI•K{OCü€ë]n›ì­ö@€«_‡'\ï²·Ó»+%¤®­õ‹¬ýò ªcj$îeû.“;_Þ©úìN—}^dÕšø+¾§Ðkr½÷åazï‰ÓGBÏ[¸&Áñ®\­5žÐÁ¼ŠÂ"*¥‡¯Óòºœú,<Õ„³…#àábºìp„‡¢M]CëI¨ØlÿͳQíAooVþÃrµ^Œç妃ƒeÜistÌÐ/ý8–ø{Ñ„r‘ë`-¯‡W}ŠÕYT¬þ 1§Ë+ä!FCUÆö»Ÿ´#êW¥-îv¾ßQ¨CôÝêìOÆ,üiÛfÛO₺yz÷íNagÑ–Õ¿Ä@e\~û-Oº®”•¡ŽAwÅ2Ü™°Ï«…¯õ‘Ñï`ŽËQ†6ÇþÍòjpÄ·þùf<+ü«›qó#Exøò'ÿ~ è¦\»òÞvßESígìñ‹­µ¥Æh?R­-سêSk{¨"ZZ6‘Ùß3Ù_‘]î§h«¶¨½àÊêÊú =©<´±„¹_biFWC’=±ÊWàÚÜ<©ÔKü£/Àm¯- ÚO³®OwäI:ïÀüú`Ù Ò‘èbÏÍ=j‰õÊO§³“(òÈ<éNUtàK)õ׺À1z¨Ú: Ô¤û®Èm¥I¦ƒ;¶ÆY ´tÕ’÷¯p†]‡‘«£ œw~¤´"Ó½ ëyêÞÈv§ÂŸ—Ìþ NÅDÓ¤ ú梬ŠF&Õ¯)¾)ªRåí¸\V¥ÊonÓZéÕp çü‹ÖUaŽdÀæÿGæ> endobj 558 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [485.286 450.63 493.344 462.931] /A << /S /GoTo /D (section.4) >> >> endobj 559 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [342.585 56.033 350.643 68.683] /A << /S /GoTo /D (section.4) >> >> endobj 563 0 obj << /D [561 0 R /XYZ 125.795 791.866 null] >> endobj 22 0 obj << /D [561 0 R /XYZ 72 754.004 null] >> endobj 564 0 obj << /D [561 0 R /XYZ 84.372 665.506 null] >> endobj 26 0 obj << /D [561 0 R /XYZ 72 330.895 null] >> endobj 560 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F18 190 0 R /F70 201 0 R /F79 203 0 R /F80 204 0 R /F81 206 0 R /F74 202 0 R /F77 565 0 R /F83 207 0 R /F82 566 0 R /F73 223 0 R >> /ProcSet [ /PDF /Text ] >> endobj 572 0 obj << /Length 1836 /Filter /FlateDecode >> stream xÚíYßoÛ6~Ï_¡§Ub–¿)bèÐ¥k‹iСA,̓bÓ±[Je%iö×ï(R²¤ÐNÒ¦köbÑÔéxwüx÷‘ÄÑY„£×;xðÜ;ÚyúJ¨ˆ`¤±&ÑÑ,JÄ$gsM£ãø#¥btrô¦ýîé+’t?rMxU‹Oñîh,ˆŠ§¤nÈxJO\Ï3÷Xdç#ÄÏo.¼ð./ýG+ê{ždùìå»' N¬C“ÇT!ª<DîŒHóiÏäˆÈÚ1'ôuŽM;>HëÃ&À7ßúà]Ë6ùB´B ³fDäetDÒBPç/åH(™@4Iœ(C|4&ãøÃHÓ8-³´ÊŠ<]Œ >I¼—Þ˜UhD ú$â‚85{éÊLÝ'EîžY>Í®²ée£jQ\›Ò5O‹Ë|ºríªpOseÊ/™›E6/Š©YáîÈš€†×HH憿ð‚º+æ1Ù„ÄÆ<¤#Êy#tã$_fe–Ÿ…9 ”ØARºâ2pp6(=²s¤9öâ–37Aà±Ýû@> )ë¤Ì–ñú;vZ×1q¬†=Ç–»=9b¤¿e$"¯Œi¹6ÿy'ôâA‘‚"‚õÃ1 iöИÃÂ!?OVî¥ÉÝ°ÙÉúøozÆdêŸ °f‘¢âqðÐ(»?èÃñ°­ÞÑïyÞ³…]!AäôFâ|Ã)œîæoÂÑ5Ó³·=Dª˜ÔOéNÇ"[K²g9ª„=ÎVî€ âЄï¦ºOŽ‡8\s±ga]HbÙ="| (mŽ0ï˜DŒˆ­k”  @€1NToÁe!7â°ƒR!üljë6 ;!b'UÀ ðEîzžYP­Ò4¨§0úæ¬ëôfpÕcõõQ×bá/’¬+ãåVfâ»ýûõZïx:o/,OaKÐþ™“¶»½ÈœÓ¶¯w§¹ñ>PÆÇ'¾ã,í\~ػ煦øášÇ'ë Í¡_s¡¹öè¬ ¡IsÄHú3 Ëö_lý endstream endobj 571 0 obj << /Type /Page /Contents 572 0 R /Resources 570 0 R /MediaBox [0 0 612 792] /Parent 458 0 R /Annots [ 567 0 R 568 0 R 575 0 R 569 0 R ] >> endobj 567 0 obj << /Type /Annot /Border[0 0 1]/H/I/C[0 1 1] /Rect [136.128 478.456 481.764 491.199] /Subtype/Link/A<> >> endobj 568 0 obj << /Type /Annot /Border[0 0 1]/H/I/C[0 1 1] /Rect [183.722 464.825 543.547 477.65] /Subtype/Link/A<> >> endobj 575 0 obj << /Type /Annot /Border[0 0 1]/H/I/C[0 1 1] /Rect [71.004 451.625 110.578 464.101] /Subtype/Link/A<> >> endobj 569 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [328.021 329.865 336.079 342.516] /A << /S /GoTo /D (section.4) >> >> endobj 573 0 obj << /D [571 0 R /XYZ 71 791.866 null] >> endobj 30 0 obj << /D [571 0 R /XYZ 72 668.581 null] >> endobj 570 0 obj << /Font << /F57 146 0 R /F18 190 0 R /F59 148 0 R /F70 201 0 R /F79 203 0 R /F80 204 0 R /F30 574 0 R /F81 206 0 R /F73 223 0 R /F82 566 0 R /F74 202 0 R /F61 150 0 R >> /ProcSet [ /PDF /Text ] >> endobj 578 0 obj << /Length 3304 /Filter /FlateDecode >> stream xÚí\[oã¸~ß_áG±¼‰—óÐÅ^Ðb»˜)Ptf[ŽµU$Ã’'ãßËdI¦-%c§Y`_bZ¢ÉCžs¾sáaðìa†g?‡ýç÷wßýé§XÏGŒ :»[Í$IÎæñìn9û8ç7ÁÏÉþ›æÙº,—7•jþÓ®XÔYYT7ŸïþƒÈÁHcMÌ x1ŒVnŒ¿ÜDL«yÞÂ|_ù!ÜŒuG X¢XiÉŽðÁõ!ÐI"-±í#r$˜tÝ>aF]O©»£ Äõ:ÆØwÄÝŽQΛI÷®‡Âý¡¸j:ü‚"FYÓc¢†ÃÊhÓ(!¡-TЉw:᧬^»}[ï7év“l“Ç´N·U³5¢·èÀña€fcú³˜-Œ{¤Ø²Ê}&æCÎe±Ì §’Ü=_¦E•ÕÍþÐÞÄ”#+ _ æÆýgˆÁ)ÅfQ§ÛïŒÇ|”ÇdŒÇ=J–¡‰ÂþìCX±j»¼MCâq+p–’YćU ØZ†b®= é'Œi‘ñ|UnMƒÏ«E’'[÷0Oê´¨}‡Fƒí·/I¾K«Ð2ˆ’HHX…¨”ˆË¦CRx Ê]½ÙÕÁ1c£ƒ½J÷$—“ù_ ·¢zzqSÝ_0ÂÛ²äçÿ%4.g-±uYæ÷å×[3°œ?¥Žò]åY¶´sãþU!ˆ¥HÓر$ž¦bªRˆ¡RÄúyJk>ìÌo¡1Ƶ‚!¬7Æ–¼ `LGhì¥ó~HÞÙ « –ôjo͉±@ðKFܳÌOÐÝD†$;V›î&kÕWš!Ã@V@d­Ô×fÖ8‚õvÅ >L)cÖ߀á0¤ÿÐk:Öu'sæ,0Û³CEÖS[6¿[m’džl6Ûòkö dàyV¬ÒmZ,üû´xÈ ß^–€F¶U”µýu“g‹¬Î÷¾KVÕYñ°Ëªµ{pŸÖOiêd("D£˜Ð>F.ò¤ª2ƒ“‹ÄaØ{‡X¦±M¶)th^t”þÏ7gµ{c,|®v¹!´þé6[ø÷~Uƒ¡üœy^>õ¾[é>-ðØ×yÚ(Eo%å2u h¦2Ÿõº}ÐÙU5¯êtƒÚ!4¢Ä{vC¯°1InÜ7Tå>‹4]¦¾Û}Y¯]k¹ÛÚ&uFOë,÷Ï6Ût™- ËÐÒqL¿Ó¸²~ï½~ð¿kœÍ#w• k#w•êÃ\Ö¼g¨©‡ñ×­¸~õ]º@Ш`öù8!]h/»q@CC‹c”ñÄ’: é6!à:Ä­ú!è:PĹ8``Ä9‰9$‡­ÅÇ-8íp·aàR19ëÎtMö)€OAw­N*·Ùã&OÁ;i¶++BF^S¡vk6è1hä©Õ”¾±‚1' †Ÿ²\ R9ô`F4 è{XF¢¾˜¾JÍÓ¤z} Ú‰¨b¼ä´'VûVøbè%…>ŸéEôˆð‡æ’BÚu€Ì/É6KŒZ†He`Ÿ;q cAS*ÀL6ÐÀb@¡ t¡…P˜Où-øX¬ÝJ’­ÝyÑ<‚_™ú®¶å£k96B‡Ÿ“€xRxF%[0+&ú±£´6¨Å“¾òEDD€åÖª‰‚ É;89Zveù”½? )eÈ fèÅу˜ñvLþdöU€m@º¤Žmñy¶q«Õà7ª2Ž ß"n¬&‘}´¡C·ÆmþŠ1Ô¢²EÇ/YŒi€DÀªˆcèìýÙÍ ˜dÏ›}@l7ñRóC0Ð$ˆI5Ý<×Zö÷àˆ(‡Ñ„bŽ(˜B’“ìÒÔ‹ì‚&)ɬî¶ýN2ALa¿&§Ÿ¦bÅ” ´ Éë¥aîÍ ìØF,C;„I&9•G³ZϹMÏ>B˜{Ç5µ¯aÞ~’‚00mL…øš¶˜ÎHHã(1Âý‡Â}»Â©Óò‡_ñŠ~ìw,Èå< pdb2ü’pq-@ש)~„&ü†ßKÌùlÒo6{Ãb9,Mtfò±wÆÍÓò¾N2âšo÷û`øÉa5±œÉI¡D|ž ¼Hމ™Jˆ'‰ œx¡œâ½óè$hÃAF½¦Ÿ(§¢€|E³+§¢ª|ŽÙ mDòXè³öר6¹©CGN‡ јȋÐñå¤þ@?SèõðKô„þR8¤Àð(ˆ(&|m@1˜Þ”^´Ø5¼f3ŽY>Ø1°Ñ¡Ó«øLàw üR0Ü Ôœ»J›¿sÈwפÝóA:}‘äyßÀ(gDbÌûGŸÀ8íý„§k6 û˜on÷»Û}E?ßD1&õif7²Þon݃7dþÙ·Wð ߪè© î óäùÕ5Ç0jž~MþèŤڲʞîÄóÍɱ?L1¯lRµ{‘—í›*k€¬X#ÐÁ‰“ß”-íË”´©ÝcÏàcrþVbš‰q±ÕÏ1õÔ“£žpÐsœ@kñÚ•&Hu¦è€kTÉ wmx0C¶»3‰Õ{/¡Ë²ð-ŸÚëúÄŒÃe_+®¦´{PÚ—ë,oræé6\ A½ZyÍ7Aéá¶ô ·gXƽ¯*]ìïSÏü+FŒLÕêšçp¬Øjµm"!~ÄÒŸ}êjßgE²Ý7ˆ]¥ýsÏÊžJ5íÇ(„ƒ÷&é …µŸ¢1ÂÚ Åið€ÐÔçÉ‘j ®[ý‰ÂzÑq¡Èmðšywj5¨´P§Ë Â¥>ñ¡J¡=9&!Š¨ÏuXú1·‡’'.hO„Müw$ï]. RBt)vÖÎEAêµ ãKNÛŽóq$Áb[®„ò—Å€º7Ú{ÇŸ ™2ŸzøI  c¸ h$<Ó¡J WÚoÞ(béºö>E÷éHÊ Ãàv™ÃÆè+—0¼HÐ)è±&“ó^9K¸"-ìÜ"5‘¼#ïbPp¨¡G§ßû ãt·6‹TIDˆ<ËYJˆ0™/ó(Î'÷ßæéÌÛ=“sE®¸Ÿåa§ð¦%a¹¿”I57/âÙ± 1 ½k$DºFX‹áâôi/ôct‚ƒTuœL€d†bÒók>‡c*ÍXf°ä”ÅKÍŸ‘‰z°žt8]"èn$E{5œqN\ç†éTã3tpî¸3M;ÎýW±Gáë™ì2I+>vu£N·?J‘®ûü[`_N€}öâEàÑkK§|b5Á)¦W¥=h’L®‹›¤Öj}5"xÔ"ÑéÉ盘¢H sŠn`€wOѱ¶µ¬Ø•»Ê·9'Ó²YQÓè^5²–é&-–î²|-ƒ7Ê4â‡m°"ÄäN¦Eæ¦G/²Ö©#ÅÉó/Tž¨R‘cØ­& b0«ò:§aáƒ"S0á§+ {X”yÎ%yUº–Ë$"s‹DÍ¿÷ÝŠ²n¹Z¯Ãwà¥u“ýRô·r\H;t5 ŸFà?„GÌ.e¯z‚3ªWX÷ ÞHx¸þ;Ü~}8æ¹öÖ?ƒpqLøZªg]¯”fþz&ãÊà0Jõ”l\Ëݳ‚F¹]š“ Û\¹Ï¬¨Ó‡msß ¸û^0»nÕýùáê€ùÖ^p özRÿŠçèu þm<^î÷<<ÅKrár?>Á&½ÛƒÏ­š ç£ÇËýÞÚ-þì[æÊßù¤3ƒn£~…„R— FâóÇPd9òò·ÃH=½,ôbr2( `«Næx\@˜L>dþ_Rìl*`ê.é‘zwèÕ©p}<—p¡ôKªÈ÷áƒZ£<Áž³õ—Ôúno88ÞlìÂ2>Uµ_é¿W¼¨#|‰Á8¶\zSZ¡Õžcæ<0…½E†Æªô7¥Žcáœxþ”å¹k-Ó:É|{]>™ñ6Þôs7M³wÛ ÝþûÕÕ4C£wÕ –uˆ ÔUpî¦cÿy"júõ?Þ}÷?²ðU endstream endobj 577 0 obj << /Type /Page /Contents 578 0 R /Resources 576 0 R /MediaBox [0 0 612 792] /Parent 458 0 R >> endobj 579 0 obj << /D [577 0 R /XYZ 125.795 791.866 null] >> endobj 34 0 obj << /D [577 0 R /XYZ 72 754.004 null] >> endobj 38 0 obj << /D [577 0 R /XYZ 72 622.102 null] >> endobj 583 0 obj << /D [577 0 R /XYZ 197.823 509.887 null] >> endobj 576 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F33 580 0 R /F133 581 0 R /F79 203 0 R /F70 201 0 R /F80 204 0 R /F136 582 0 R /F82 566 0 R /F18 190 0 R /F74 202 0 R /F83 207 0 R /F77 565 0 R /F73 223 0 R >> /ProcSet [ /PDF /Text ] >> endobj 588 0 obj << /Length 4092 /Filter /FlateDecode >> stream xÚímÛ¶ù{…`¨®è©â«¤û°¶I×­M‹ö¶aHƒÂgËgíüY¾ä†ýøQI™²ïÊydËi€ÑÑIQÏû£ÁÍ |÷Ydþÿú곯ž‹t@H˜ AW“AL1gaÄÅàj4_®Kýãh:œÍ²ÅM¾¸Ñ?-¯ÿ“ÌzÕ·)ÌóÕ,›g‹rXæËExq)’(ø6ŸL²BuÚ‡à/ýÕ² Oõ²‡³›e‘—Ó¹Z,$(²×›¼Z04oª~®Wj%æÆåD߯Pulß[ÿ].õ ×æ÷Ñr¾Ú”ÙøËêϸz®È&ËÂüZ“×{Oçv]Ãê‡Þ _Ü-gwY5kœ¨5êÿÇY9Ìg¶÷:›-ß\¤4«j`°ÿçMÂTJ^}ÙKFB5M ëiÓ@3Ì‘¾gªo‘ĽEçöµIÏ7‹Qõ±ÖἫíÜ1 ™döö‡'MÃt;gþ[D ¼UÕUý»Ø«¿ë7S÷K*áÍŠ”«7ð7“ììI(!ÕŸúòØ!ጣ'Ô.áO¼û×ÁËKYÐÅ%O‚Ú[hð¸Ñ?…v ýK¸£L³`ƒn ]ÀÐ#¸.¡Í)p½F™Xò`¨5 Jó(ÈpeÁ ÖXÔ«–Áóch3èÇÙ@‘@pÔdé]¾†áp6Dpø°T}´ªÍ EzqaVz3…}L+HWm†5(°ï®z€ì‘ÊGî,˜ã,1ÚC@ùJõ«×âR?mÜéC# qkÐHŠK#];=©~ïž“$b@±|1”\Ïà»p=é~‚Ê8ø-‘' D°@ ´mj,(°ò†ÓThíÚÚÚ¥ùÕkÑ„DÁ—X¸7‚ï3ƒvíÚish'õW•æ~}géÜÓ ? I‘Þ@±‹¬NCØÚk K)!‚ ´`’T4Uµ…#ª¬ází,KCÎüàW@K>ŒXH£Çœ±Ð@—6ľê³{œÃ=z_KgwGݯ‘²„· %R/ßâòƒÝëß™éÓß; '„Ÿ=!ÐꂦÞ¶_o6G ïj˜Ü>…ˆMÏ.…ØêÇ£Fœýjäy*€ð¾…ëï Ör«F ÕÐY[¥Ka±´¡ ­5<‘B8²ËW“z[›|õ$XŸ_uÍ)±á«ChWÐÎjÖÖŽAá|?GM•~ÙB8gÎÂccؘ9=#G?zƒ…—Ic Z3Ž 0»XétãȨZ_B²ýh,âÁ[tUsCj‡_íuYsr0C©væ‹û ÷ÉÇŒ÷pßR€ã~r¾¬êך¾ÆFf›Â5í«¢ÕbÁôWµ#h§~Ϧ2¸ôœx5Kƒ×`wÝÀõ® oË[µÝj·;C„ÞŠŽéy Bˆ`…›[h¯ºŸYrµx¨1Æ@€UDðá Ðj⥿@&ÁçÞÆ['õ^âücË©M12dtÇC¢ÐEs¸¦å®Ãºúî:F¦°Ðø8ð|^üê@Œá¬îvg\HVë„©1…[ñ‘cMq~ÙPž·²Ëuýýb¬)I0W° ÀÀÊLXâõ=€ÀbÑ0ζ¾>4×Ò89QáẆkV[CYgO¤ÚÅNЖjgbE¥É²Ð䦃–&ÄÍ3ìÂ&Q8!´Ú2qƒ%–Îæ¸æ©ÜÙœ,Dƒ²óß`i óFÈTÕÞcmÎd{hWÐf ÷NÏÚúõõÜÞÉa{Ed® =‹´™ÚbQâ½´'D_ë[hWÐêž‘ã›]ûû`ÑâY7 ­¡ÍAGC5¹“|’›h2×Ýuç¸ku$wû˜/—˜5l1G±UDÉÓ¦„Ão½ÝðO°Ì.9¦AC}„šŠ¦@cìD°Ì‡Îb'Ð+FŠ:šo‚˜§‡½9Ñ#vŽe—°¿Ì ­CþFÞÎ*¢P~õFÜ}Žï‰ìdíÜa-3Ã`ŽÍNÒFÂF·r—’?žxõvµƒô‘, ÄÉJ[déYž`½ØñD{Ù ¡¡Gó|®VÑÅ¥¢mú:D±„QÒ­Òž¬ƒõÆõCÏÆõÓpâ8 ‰›BÀÁŒa“ uZ†rÇ¢G⸄lœÿ¹„cŒzû„8¸:piÄàºH”:PµßCû ô_9=?AûKú`ã'h¿…ö™·§ÇgByO(£8 ŸI B¡g/gÂÙø~õõ$X9ÔxîtjÊ—X_g{ã=1?e ®×*õ%ŠJežU¹|?pŒõ¡8O_s€¯Í‘áKavn­¨7½Âïô%è1%ˆ*¨¤Ò*ñ4i…ôpt0K–ý§ qOlPDAk°¶åBðêËÕ9Æ3'ù b±HPË¿‰ÔØ„ïjk•MØ8=%fbëÂÎ"œB|ÖùáÚ¿G{%mîÒðMRÒر“­­gФÊÙþk¸ÞìÙÕ–píŸ0b±™lò#K±ü6Ì—™Ö&zEs*üîm=×Îÿó¾rÐøhÒ"v!Æw•7<˜ô$ræÄ5Le„êábh1%Òí²F,ÄB§¶Ü“ë/±ÜãnžË‹½ðV™Z£Cˆ…ô%F:u]TÕ#°,úrð1GL¤³rp4®¿­¬ Øî0Þ´ãõð@“Ìh7ñ"+'·jíD‡dN›Æ_àpŽž„‰ÈÁ·&riLäy×^[$fµ:C"¤îÛf|µ5¾ ¾¤ 'ašôÞ·>‘ ÛK%+ŒÜ:½-v.º¨)ŽYZ›#Sú¨›ljUìE55ÕX a*yiþÔQìÇÞ‰Ka¯#ÆW{uÇxXuŽ‘t Ö“ð$A¤Â{-¥ŽA;סáÌ›c°ˆc`Æ6DáÄ…-X¾uj;Ïë‚Ž¼uìúk)þœä¡&#ä œÏ$NDðºa°jlóšœEÁo„qoŸWîX Ê3°È®ÈïÇi5 é9Û“åéìÉ–6Ô5i~÷F¨/pè1í‡ç‘±³à )à JJ±Ó¹uÐþ v(‚xÝÌ[ðôFˆ#ÖA¹)5èÏêµ­‚^ìå¡êR%ÍêèVjç]£Ø¶==àx‚'Ë߀· sñÊÇÄT‘Ôê^Âp¯°V‡n®8ò¶¶ð>år0[ççÀN‰¤´‘[62Zè S*´%>FN¡%:­450»8ÂE¤]1’š n½Þ9.ÓvÏ#ý¯÷ÐúŒSœÚ",íGæ:ïɹQœô¦¶;ŸÚ"ÔÔêÈñ‘šãV¡æ(D]ùm·¥Ì^)³a},3'}ë æ†9Ûdf]÷­ÞHhŸx’à;7Ù.Ü\—æH¤½vôšígdHe¡@Äsî˜XÙY+ðÒø&A~âò hŸƒ¸ü ®ÝJ0ú×oLÏGSñ…÷¤ïK-¬Èn˜ÉûuOÍ:±ä7NzŒòõùÕ®¨üù”‘ ¸e$bï(»oâ5LL<±9€02v¡Éá|`!¹SHçP­ƒóOTëÕúDµ>jªÅ@>eÚšM}©Ú7füÝÅEOäBÙ“r?ºò#ïìTGy«wv[,UÚRäÆ?M_#jÎö K9Üb4ñÅè4ƒ£êê*Onb¨®ë}L•9!Íxû­I›:~KÄÐ1+Íj¨aíŠÙÂñëõñ㶧ÎZgôþ<ƒögôò',Å+2ê ,ýÚ¯|ô?⓾D øhCÈÕ~\pǯQ 6ðãŠÀ&%ÖÒì‰s]`Öç‹Qc~%1±`™#¶skÚìֵËcV”ô7 ½©sÑÁ˜%ê˜'r<›Õj£ Q I>|ØY]5iç°³úzç°3ç)ߣa‘ˆyÒå&í“VWY!'‰pém˜Ñ>Ï ïàdÓ6ÝôäÅŠ„ä2€A-FzIc“> endobj 584 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 668.883 51.396 677.323] /A << /S /GoTo /D (noweb.NW2i2LRH-1oKncb-1) >> >> endobj 585 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [148.611 666.246 160.574 679.311] /A << /S /GoTo /D (noweb.NW2i2LRH-1oKncb-1) >> >> endobj 589 0 obj << /D [587 0 R /XYZ 71 791.866 null] >> endobj 42 0 obj << /D [587 0 R /XYZ 72 754.004 null] >> endobj 357 0 obj << /D [587 0 R /XYZ 81.963 669.969 null] >> endobj 590 0 obj << /D [587 0 R /XYZ 72 667.242 null] >> endobj 591 0 obj << /D [587 0 R /XYZ 83.504 669.969 null] >> endobj 592 0 obj << /D [587 0 R /XYZ 83.504 658.014 null] >> endobj 593 0 obj << /D [587 0 R /XYZ 83.504 646.059 null] >> endobj 594 0 obj << /D [587 0 R /XYZ 83.504 634.104 null] >> endobj 595 0 obj << /D [587 0 R /XYZ 83.504 622.148 null] >> endobj 596 0 obj << /D [587 0 R /XYZ 83.504 610.193 null] >> endobj 597 0 obj << /D [587 0 R /XYZ 83.504 598.238 null] >> endobj 598 0 obj << /D [587 0 R /XYZ 83.504 586.283 null] >> endobj 599 0 obj << /D [587 0 R /XYZ 83.504 574.328 null] >> endobj 600 0 obj << /D [587 0 R /XYZ 83.504 562.373 null] >> endobj 601 0 obj << /D [587 0 R /XYZ 83.504 550.417 null] >> endobj 602 0 obj << /D [587 0 R /XYZ 83.504 538.462 null] >> endobj 603 0 obj << /D [587 0 R /XYZ 83.504 526.507 null] >> endobj 604 0 obj << /D [587 0 R /XYZ 83.504 514.552 null] >> endobj 605 0 obj << /D [587 0 R /XYZ 83.504 502.597 null] >> endobj 606 0 obj << /D [587 0 R /XYZ 83.504 490.642 null] >> endobj 607 0 obj << /D [587 0 R /XYZ 83.504 478.686 null] >> endobj 608 0 obj << /D [587 0 R /XYZ 83.504 466.731 null] >> endobj 609 0 obj << /D [587 0 R /XYZ 83.504 454.776 null] >> endobj 610 0 obj << /D [587 0 R /XYZ 83.504 442.821 null] >> endobj 611 0 obj << /D [587 0 R /XYZ 83.504 430.866 null] >> endobj 612 0 obj << /D [587 0 R /XYZ 83.504 418.911 null] >> endobj 613 0 obj << /D [587 0 R /XYZ 83.504 406.955 null] >> endobj 614 0 obj << /D [587 0 R /XYZ 83.504 395 null] >> endobj 615 0 obj << /D [587 0 R /XYZ 83.504 383.045 null] >> endobj 616 0 obj << /D [587 0 R /XYZ 83.504 371.09 null] >> endobj 617 0 obj << /D [587 0 R /XYZ 83.504 359.135 null] >> endobj 618 0 obj << /D [587 0 R /XYZ 83.504 347.18 null] >> endobj 619 0 obj << /D [587 0 R /XYZ 83.504 335.224 null] >> endobj 620 0 obj << /D [587 0 R /XYZ 83.504 323.269 null] >> endobj 621 0 obj << /D [587 0 R /XYZ 83.504 311.314 null] >> endobj 622 0 obj << /D [587 0 R /XYZ 83.504 299.359 null] >> endobj 623 0 obj << /D [587 0 R /XYZ 83.504 287.404 null] >> endobj 624 0 obj << /D [587 0 R /XYZ 83.504 275.449 null] >> endobj 625 0 obj << /D [587 0 R /XYZ 83.504 263.493 null] >> endobj 626 0 obj << /D [587 0 R /XYZ 83.504 251.538 null] >> endobj 627 0 obj << /D [587 0 R /XYZ 83.504 239.583 null] >> endobj 628 0 obj << /D [587 0 R /XYZ 83.504 227.628 null] >> endobj 629 0 obj << /D [587 0 R /XYZ 83.504 215.673 null] >> endobj 630 0 obj << /D [587 0 R /XYZ 83.504 203.718 null] >> endobj 631 0 obj << /D [587 0 R /XYZ 83.504 191.762 null] >> endobj 632 0 obj << /D [587 0 R /XYZ 83.504 179.807 null] >> endobj 633 0 obj << /D [587 0 R /XYZ 83.504 167.852 null] >> endobj 634 0 obj << /D [587 0 R /XYZ 83.504 155.897 null] >> endobj 635 0 obj << /D [587 0 R /XYZ 83.504 143.942 null] >> endobj 636 0 obj << /D [587 0 R /XYZ 83.504 131.987 null] >> endobj 637 0 obj << /D [587 0 R /XYZ 83.504 120.031 null] >> endobj 638 0 obj << /D [587 0 R /XYZ 83.504 108.076 null] >> endobj 639 0 obj << /D [587 0 R /XYZ 83.504 96.121 null] >> endobj 640 0 obj << /D [587 0 R /XYZ 83.504 84.166 null] >> endobj 641 0 obj << /D [587 0 R /XYZ 83.504 72.211 null] >> endobj 586 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 646 0 obj << /Length 4930 /Filter /FlateDecode >> stream xÚí]ë#·‘ÿ¾…àmlõòý@ÎœÛHÎg$ö v³ 4iFÙyE£±3‡ûã¯Xìn±[l‰ê¡vzr‚áÞ‰Í"‹U¿z°Ø"£Ë}÷†”ÿR¸ºÿ>ª»ÿzóõÙ›wßJ=²…ULÎ#¥ !ÄH ^!Gg£cÉÞ~<ûCÝrTÐÑÊýéoüÎõ#ƒ~¨*8cÐ{ø·„žóíèÄjËÇ×o'\ŽïŸM úSÜŽ¿ÊÒ‘ðã²ã»·AÆ— ºÇÔø/D’,#`6‰'ÐTòTª¾ñSbK6þßä>É}ÂPižµ&Û}Åô¥¥,“RÊ'”VVÚÂ_ƒ¶0FÆ(^6ïs‰úW¹äâ³`†é¥›2ŸnÂL¢›,Y7y²n²dÝäÇÕÍgôõ.OGÆiØšEÌ/¨¿ A^¯ñš äç…´{ã¿â"±,„%E†‡€/`ÇÀ özìxF¬Gg€Xa£XÑù€&Q±í-r`ØÂÿ¥°…Ÿ°å„-¯[xlQy±¥÷8ô00–cšìq¸hV™úHU#ØPPxwý^}0x…÷Oå·I½QbÆ_&JŸ@2ÝK?ƒƒÐ8]~{…Ÿ?áý}rÏsiy¦¬(#Ë%^?åFt‡nƱŠãzq–Ê*c“–+U4ÚÁXf#¹HÖ¬<$A™? {?¾…?2u Âñëäy€Aû[¼.àÊÇßãý¯÷x½®?áãÞϵ—º¡8ãûe.Zö·Íã¿*ñß ÿ%¯€û ¸ï®k¸Êñ#ÞûÏoñú…˜fˆ—‰$`PXìøg¼Ÿâu]nÊÏçØæ!—ãúKòzÉžàê}®³ö™ä ¬Ñ¸ËÅéue£¡ëyžNJÁ‘Zè:G§`Ä®PÄžðzW/t«à“ið‰¿¿©ZJPðT'Ѿ†Ä¼!ÍrI¡cä#gÏU…\U™"g`ô+KØëS | ”ûî"d5=ª~¶ÂSoÇ%ª·Dõ–¨ü•_¦:TšæÛÄTd Mˆ1 ‚ûôˆ1‰ðgO)Ò¼)EvBÊWXvðÊ’fÙÛè-ãf€ÐI‡‘Ul ãàƒÉÍž£èÑ!çf=j3=~Ÿ(¹ÉÉW¯+ûç4µKRûW&R;ð>ùÙ¹³hOSú6g.•¸„û1«p'~@Âìs¶©‰^)QÔŒKrí½pÿ Þÿ15•K\*—Ë|yU5”º¤ª„v^&«2uú~X ã\/ÊEr¢OÝÝ"g²ñäT¸g‰~8NL³8BLãÐV%£­M6õÔÁéÛÜ9c&“-7ƒJ$1ÓÌ0‚`ˆ/ 4*‹ã+_×±‰÷Ã96±— ³28UqŒô†W ÷ý“½ÅF½ÆóïOçŽq~àÙRøâé$XR/'7ȧÇ<:¢\/ÿþT/ÿYjZ½Ä¿0§np¿ñM’ý4‘µ>\™ÎV±¦XžpÐ5œš–™ïk¼¾dÁÓúaÃ̼ '³N–™·Ü±`ræM%gÞ²6Ïμ™äÌ¡¯¹ˆ²4šœŠ(OE”§"ÊSå. ‰¼lÿ'ŒRB¢±1yŠªö‰FZQHoMH}…æ‡Lü6~‡Iv"}÷Ý÷yž»O‰) ¦RG›¾‰$YÙóyöž«J†AÞUÖð΃©Oi2„àóä§NGÿDÚþ¼ Íz^+0ôNW¦¯+Šþ3Þ}XýŠÊW¿¢å`lÒ%Ú¤iÅÕÈk™+üduäâP-OP²ÿ3ÖãOñê#÷eð‰âgxŸ)ó›Á¹Jð0¦„ÆwÁÄËô…»÷×%~R±% ÷My$ø¯¿àÁà9Þ¯ò.K^xDþŒòE® :kàÏŽø+~ê21ô)H$…Ÿû÷ÜÕW‹Eì_ð|E去¥/ò&¯¢ By:-“±:’¢ÅU¹L -KOÿ™ü¶Ìü‡©Î“yþ›¬ë˜ä&Nò8ìÉEš<8³ùW¼2¼¾Ã+ò|óa"3 =ÀGË7°¹ø—î"Ä-J.©ç4†åJ~¾ œhZfWx½ rŠëò“L©ËŽS~½8BùõyrŸ¿IîóiLs~:T€UÆZæÅÙ1¼EÕ4ƒºuíkT•)¶ÑÛûÃêhûÃ0¿#ÄƴܶZâõ¶žÄ PæŠÑüÛ¡vÊ9,×9‘´´m’PN‘¿·È߬±cÞEƒ‘þ*Ù5y¨õÀÇõª|§Ë2ÐË\Ú°¨5À+ P ˜–\pH›j’=Š;r¨«–1 Ù)°½MÄPÞ_ÖU ã«^ìÎj“ r«w \ ëw<²má3›¬ñŠp”:|«8gX‡’üjóÄ—ß ó]+Çxo5túBu(‡¾G#ap#xÿg¼ÿú°ÓüóŸæ`ùJ_ÌP µuXâ`ÀJ"ðUÞæ*W “ùµ/]xÂû{¼ÿ[ ›‰r-óEG)¦ØäôVîÁ”õ1±ÜÃé ÎQÏàˆcäÀý¡¼'[ô©äæÐŒ$}Í{„C3f G Ì[}‡ªûˆ×›ú¾zÃo;ͨ³ªºöGhlGÙ– ʶlP¶eJ|zÈ%ª>z_åÝ»( *«êÂjà2ؘ—ï”Ï´É´Þ€²Ê•>omÉreUêÉ"wJl)ÓòHçæê‚.‚ûYÐr|~‡×Ôý• ™•ÛÒ-ã@_àÉ‘CÕ½ÿ½˜»Æç‰ÅØÙ~ë&ö†­§¤¸ ~³$La,ƒÏƒè:g¼óRÇT”špZæŸ;¡„ñïoî¯ç7óÛõüÂðûå§ùõòêî®üûÛÇÛÙzywûP˜’ÂK]ßd4aª’únÏ®@†9 êâîúúî—å­«Or™ÚéùuùÍüöñf¾š®çþïéõµ¿q[²³ÇÕ †rýTäs[áø\»ë`|Hªþ¹¾š®ý“³é­ÿè¼Ôã-ê6å…Ö{Z.5áÜ(+h-R@õüÿï?üçóÅMŒFÒÒêémÉ¿õÕ|¹ò·«ùÃýûóÜÿý0_û›»…ÿ÷êé~¾ºŸ®¦7óõ|Urœr’aD¦’¿ÎbƒQؤªQ•“oÎJÕ #Í`¶²ÐÆŒf7oþñ…sªü·Á½ÿ²|ˆ»/FF¤á›Éè¾öÜšÑ/Øîf$-qO]~zó§­-(±»z€¶~µ|Õã9°Ž›q}ö¦½†FOGÔ§@ž~ ‘#ÀÍ`(g1Àj£ cõãAºMöSª¬x“òj~ ð²Z*6 ‘@ƒòjÉžÊ&b¤ «I¹¦ÂÀ„ -E ÿA?º`x°iUÊo´‚[NtEíÇmv0É K=ÙѤ#*”h-Ä+ñæ9‚mt“T\µ%\Û¦„sb6ô5mÆŠ‚ ч¯¦Ý kðõï1b ìHMl!& iXtíNjH|µ½X‚[ÀÛw±Z#„, ÓÍÕêZ[([Mdh’‰Bð>r´éCK ×…²¦ÖT}(Ìi÷0,JUÈ¡ªhgⵟùwÓLJ‡ Òl T"µÂËéí îG»)ýµ‡´ˆ „+¨7æ©['y´I:É{éd“ÿ[:ùe|jFÒÀ<óº«Êñ³ªÐm°0µÉ¦U Z=2{nÝaY±=Ý–îZIXC ¿¥F–¿’‹Le­E(:$W÷$a!Kù>B+˜†Âv€Á[HkŠ0óPo+é(V]“ ¢Ó`\s-gûic Bmr|‚•â<ÿç})d´ÑÐÚ2t¬úe$n“$˜ðIØtRv9àLABniI8·f[I$4Ð L²‚ÜrcšZÂÝ0@è Ñ$ªoÖ°mÙÐÉBwk?-ÒªªÒû›ÞÜÔ4(¦±¡“Òëâ ®\΄u ˆDŽTãØ-;qÕ yn¹qBšµd×:cº¶ßŒî0Ü90º¶Û!F/b²ÏÝèªÈ&êº[ª·yѤèp¬Špw˜þƒ'5Û üég™~hÀ€Äîahù}¦_‚¾è2züi>»b;L?²¦¿MxòðÇéj~±Ãü÷£ßP ªh¤[I ]«vù„Yhƒ‡µá»ÔÔ±mdš6­îŠâèJu›P¥Ç³»‡«ˆv2h ’Æa欩žM”2€ü`[Q <ÞKÕœt«G szÚé}p._œW“ }Ö‚Ü:Â.JèvštaLˆ¼"îY*®¡F[LI¡ÿ'È9Ë‚EáÉe µË´lú£ùÁÆ¥áVy㢠]Y˜O¥u ài ¾Ë©ªHvŽ½î*;ßa¥úAI‡• ¡lVêàÉE- ÙÝÅ~+ž5ƒ9Tp™-Õîûéýõt6ßa§úî°SÛ¤wǨ½‰Ú¯XáRa‚ ÞÕV„Ú u‡òÃÄzà{-ðPH–¨üçÑìqa@àµâûcîc•‹¨f,"[±ÃX„6Cè°‚̾Ó%w™®¿G ”€Ý¡*ED–·"â†u§Î4%á¼c 9ð "[í?T±Á¶LpšÆî·žçñ,j—ét‹4-ÓÉcÖL¹Hb™Û‚B©éL"Ô}éBÞmñ /(cØNkû,‹WwE@×lgìY[¼~¸ÔañBP‚Å;xrQkEvw±ßâ¹ -%vƒqtöY<9Sùg;l]?’¶.$úÓúñb~»þõ[-üÇ›²H6nöú£‰Í• `!£@H4Lá"¥„HNÀl€–Zƒ“®» QXlvG¿ˆ&J¯¥0f‡5)÷Ài§€‡mÀW50èt[«ƒ¨Oû(ƒŠq®»? óç`Íôyå.ŠB¶Ç.pnvئƆ”ÀòÈ|«Ó\œ6kauÜðA@#û%kÝ*&q–Ààºù*q³ø*#Á`#>³q€`Ý-ü7¥’˜2¦v¥©7};P¥yïc4aqÀ ríœ;´+Qí2!lׂW}s!X°NÍdtÓ)ÎâÒ\…m·®½±/)k¦˜i½ô¾¾Y-ºq«ÉxÀÞ$ºº+«s«Bânî7Ãb lÔ&Zÿ1ªKÎO2µêá|·|ø–CÜß·vuÓš€–»h¶éFÂ<;\ù(j‚Wç"g^¨ªà¾³œL¥øF5X¬ã0 …=ȧ¹ˆš°Ð5ZwãG?IiZ§ >Bqým72L2¦Öšèg!ƒtN½Ù;Ò÷ƒb…”%f~w¹|X/gÝèЋl¢„ñ¡ï0‚䑘ëw3RïÝÃR²àjçþ5„œ”Ù„-,€%Í›êÊ»6°¾ˆ–4 ¿Ù½j¤Ü6 u9–€5Ñ›hh ×™Pˆ+/1ˆë~(;tÿPŠ1½•Ü¢úñ£#jºê8¤Ñ !|g. W¨ªÊ‡6ßÏ“þ?—Ý•µ endstream endobj 645 0 obj << /Type /Page /Contents 646 0 R /Resources 644 0 R /MediaBox [0 0 612 792] /Parent 642 0 R /Annots [ 643 0 R ] >> endobj 643 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 297.217 165.939 309.322] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 647 0 obj << /D [645 0 R /XYZ 125.795 791.866 null] >> endobj 648 0 obj << /D [645 0 R /XYZ 83.504 755.001 null] >> endobj 649 0 obj << /D [645 0 R /XYZ 83.504 743.045 null] >> endobj 650 0 obj << /D [645 0 R /XYZ 83.504 731.09 null] >> endobj 651 0 obj << /D [645 0 R /XYZ 83.504 719.135 null] >> endobj 652 0 obj << /D [645 0 R /XYZ 83.504 707.18 null] >> endobj 653 0 obj << /D [645 0 R /XYZ 83.504 695.225 null] >> endobj 654 0 obj << /D [645 0 R /XYZ 83.504 683.27 null] >> endobj 655 0 obj << /D [645 0 R /XYZ 83.504 671.314 null] >> endobj 656 0 obj << /D [645 0 R /XYZ 83.504 659.359 null] >> endobj 657 0 obj << /D [645 0 R /XYZ 83.504 647.404 null] >> endobj 658 0 obj << /D [645 0 R /XYZ 83.504 635.449 null] >> endobj 659 0 obj << /D [645 0 R /XYZ 83.504 623.494 null] >> endobj 660 0 obj << /D [645 0 R /XYZ 83.504 611.539 null] >> endobj 661 0 obj << /D [645 0 R /XYZ 83.504 599.583 null] >> endobj 662 0 obj << /D [645 0 R /XYZ 83.504 587.628 null] >> endobj 663 0 obj << /D [645 0 R /XYZ 83.504 575.673 null] >> endobj 664 0 obj << /D [645 0 R /XYZ 83.504 563.718 null] >> endobj 665 0 obj << /D [645 0 R /XYZ 83.504 551.763 null] >> endobj 666 0 obj << /D [645 0 R /XYZ 83.504 539.808 null] >> endobj 667 0 obj << /D [645 0 R /XYZ 83.504 527.852 null] >> endobj 668 0 obj << /D [645 0 R /XYZ 83.504 515.897 null] >> endobj 669 0 obj << /D [645 0 R /XYZ 83.504 503.942 null] >> endobj 670 0 obj << /D [645 0 R /XYZ 83.504 491.987 null] >> endobj 671 0 obj << /D [645 0 R /XYZ 83.504 480.032 null] >> endobj 672 0 obj << /D [645 0 R /XYZ 83.504 468.077 null] >> endobj 673 0 obj << /D [645 0 R /XYZ 83.504 456.121 null] >> endobj 674 0 obj << /D [645 0 R /XYZ 83.504 444.166 null] >> endobj 675 0 obj << /D [645 0 R /XYZ 83.504 432.211 null] >> endobj 676 0 obj << /D [645 0 R /XYZ 83.504 420.256 null] >> endobj 677 0 obj << /D [645 0 R /XYZ 83.504 408.301 null] >> endobj 678 0 obj << /D [645 0 R /XYZ 83.504 396.346 null] >> endobj 679 0 obj << /D [645 0 R /XYZ 83.504 384.39 null] >> endobj 680 0 obj << /D [645 0 R /XYZ 83.504 372.435 null] >> endobj 681 0 obj << /D [645 0 R /XYZ 83.504 360.48 null] >> endobj 682 0 obj << /D [645 0 R /XYZ 83.504 348.525 null] >> endobj 683 0 obj << /D [645 0 R /XYZ 83.504 336.57 null] >> endobj 684 0 obj << /D [645 0 R /XYZ 83.504 324.615 null] >> endobj 685 0 obj << /D [645 0 R /XYZ 83.504 312.659 null] >> endobj 46 0 obj << /D [645 0 R /XYZ 72 289.302 null] >> endobj 644 0 obj << /Font << /F57 146 0 R /F55 149 0 R /F73 223 0 R /F61 150 0 R /F59 148 0 R /F18 190 0 R /F136 582 0 R /F70 201 0 R /F74 202 0 R /F33 580 0 R /F133 581 0 R /F79 203 0 R /F80 204 0 R /F77 565 0 R /F81 206 0 R /F83 207 0 R /F85 686 0 R /F82 566 0 R >> /ProcSet [ /PDF /Text ] >> endobj 693 0 obj << /Length 3546 /Filter /FlateDecode >> stream xÚÍ\Ýã¶ß¿Â/E´A¬#)~Hmó’â$M4·( Ü]­­]«gKŽ$ïÞ¢èÿÞá’)Yë•m&Ù‡¥)šâp>~Ã!‡^2»Ÿ‘ÙwWÄ~~ssõæ[‘Ì( !Øìæn¦ØLñ($\Ìn–³÷ùõœB‚¿•›mÚä·ù:ož®çLÅÁ7Yó˜e…yø1ÿ”­óUY.kÓKSù¾¸Ëª¬XdæñY³Ò>ÞüÔÕŒ’0! ÕÔÉlÎdÈ5ÄoVö•»r½.óâÞ<6éíÚ~³Î릥·^›Ê¶¬ë¼ë±(7·y3/ Û±¼kßmgl©ìŠ…î6˜~ÞŸþÆL?Ôó¿z{sõë…y“FhÈÄL2Îf‹ÍÕûd¶„ï~˜‘0JâÙ#öÜÌ8e!P]ÏÞ]ýóp™D!Oø±! w žØ!Œ*{¤z,†ÁT(@Ë(Ò< jé€1F%räeZŒ±Ñ¤?­·ŸÓE3BXð0‰'Â’„q$„¡ªÉ“X,BÁØ@ évŽI:b*Œ¥ÊQ¿ªOù_×± Ò*G<¤0ߤOY}8ÎÀ/ y˜ Î"ú³©²û*°æ¨ 1"ô ê£pb‰¸ ‘fs¨•æܨcJ1~—î@ˆiq¹çÑŽHÏ”#­B—ö¿ Ö QÉB)äËD[Ðz!š$!Á¿D´…«š‘‚7xü"Í•>Ì©C¥Kt*O¦>Š‡èø/BJF¡$Ò¤¤ ¥²¡Ç»l±š×¿îÒ*[ÕÉôGPrÏ ÄåáJ¼mQòÑ%>hv(yf‡Ò¡Ä%: %؇c✨‹PÂ0ŽØJD*¹qÌñ•çdâ#9g â2p "L¼µxóî,þ…ywïCÛÅ»D'YüRsÌ•ƒÖ/±øH„ŠRO%¡Rv?ü®Ù-³¢ùâZñÀn'›#ÆòOϸ{WGܽšñ¿@³3~ŠïŒß%:Éø/P•c¹âø/?P —á‡jÎ=Á‡ém¨•âÛª*«ÁiËóè9u#à9‹‘g—‘c ‡š-Ö^ ÙbÍÉj/ì æÁF:¤¹4·Uy›7ÆH&¡î|3qÌ]¨Ë@GDÃBé2$ ãXÙ«¼Ïë&_LFÍÉ3ÍyÜ<ƒ—›c¸ñB´Î D[àø Ù!çšr|ØJ—èlåä\`*ŽÝËãC  3‹b3‘Pı{\cÕÏZA#æÉ\ƒ mV¯—S„1 ªe-ºy(lJqÑ~qjjƒl÷cè§ð¬Í˜óV$ÐPƒÄÖYÝŒwïÌš·† + â&«rÔÕÀÌè† 5`ÑtO#âs&Kâ†0•…iQhíØPn¶»ÆèUg"ÓõDézÝ‚_E"øÞ¾´­òb‘W_¹ó&qšhZ¹[Ûªñ¦8¥lÄS”Û¬Jq6LHQ] æXüÇI˜Rã!±Ö˜¼- 2ãÈu÷^•Ú,ª©£tŸGûåbUÖ¶ª¬?ÁÆòeÖ’¤¢é³î“ÂQ`”·=`’oà 2ÜtðÞÓV0 ï'š„Éžfþ0Š\™€„Ì>ÍÚÚß gÐ_2‰œÅv 8C§d‰ d¢ƒšêÞF7?£t„3†žH,Zò°Vpì ”Aõ– –9¶—X/zdfïçRà½m é,䆪©þŒ]À¹>àË)–ÚïÄN]÷ØRv“PöÊ!c|„H'"Á××HsíLíÖ¿ëhH;JÝ–* >A`¡‹m¢4ÐÁ ¶×ó8 ÀÑH†M’âPÙh›ÓÀáÙ&4Ȁ훤&c8Ö]7ôÍu@£iNWªÕHO©¬'†±Q^ÎÑÀ B¡UaT‚ÕQSP1kMáO8þ­5Š½‚*H: RŽ‚ óC2 æ^‚ðˆ)qiyJGðp.IXöâŠQ\àï°tÛWNÚré‡v¸÷eðéñ¯Q!qû9.¨‰^U;·žä¯ÄGRw|zÕk_9šò z‰«ÞõeB•F&âê‹<Ó˜ˆ£CµÞ~šW°ÁŤÎ4{FûøáR]cØäÉqeˆGÜσ™&!jÃ/Á­C¦éãÉTI6)0Jx:eæ˜Á/h ëo°4õ/§ c x›¾Øü‚Bf(po¶*|ãg ƒüŠe…åD]Å­Etü}9Y*[”b>Ù(&Ž ±ÿiú!Z?Z”ÆL ñ+/k”|kxÁG} ƒ«‰/LŒÅ=-@àÉ‹KF—³ g<ÍÂíbâ‰YØ¢l%–)–…¯¡Sœ)y ÀON‡Ä$]{U;žEvñoÐ݆ۧHKgSXÙO»¯¥û=8»“tt6}¿ ö+çz#å×;ø•8Åßd÷&q¸DCA ‹œ h ¶p¹ ¶%Årƒe†íM[ïÞª}M.µ»º©Þ_ þìEïÉ`~2ŒØ`‹CÉó0ãø¶?³*=MÂ' úµ¯½Æûßa+ǘ9b$Ú%‚,î'Ûäô½Ç©!¡¤~Âô Ü“8øx™Q·yêרϞ{à/ú—ø ˇ®.УJ¬ýøÕÝw„Ë:Ç0  ²k_‡<§:îI±æÖÙò•X>`™c¹t‚Üå4èP:Ü9]àÈî¿àídyáœÃm°Üv'Fq/F+ìé³ z9ŽC <°âÖ÷íf{&¬•X_N¶"”¾ð$}Áœ`6±‡ï&˜}èBZ#.ÓžØööˆÚÓQ´ÿÍd¨viÓ ÇI˜„I3’Ì:ÛF¯$Ô0*Öäj,·NËËKc¢ηk,Ÿ¦¡ôçjšûózR×'NŽ.~&G—X<Ô^SLæ|xãÑaþž‰› ¸t`×8ëqnWå©ùd-jæ)›\~ïñ˜«3—¹ÄÛ]½Â2År=U$ö…*ƒööܪ²‹¼–…ß“!cg9‹&¨èl÷Ì_‡{¶·sõÊïIn)JéÁ JsšzŠwMp´s¨ÌITçÎå£ÔúVOaTî¬`¦Ü9±}í+L|DX¬œØÛoó[\8¼‡5ÔMÐ6““¢“ ûÝ—ÈÈ. Ÿå¡rŒ6f;oy8ÛYˆ?x‡Ý_ÉWs|õÎjÝŸ%›ü7÷¨˜ML–¾œÖÚÙ®˜{ßb}gOãu}Ñ$qn×$¯ŒC’t×EÚÝçÝš„\&ÝuÞÉ·>ò$ð—œ|í“‘¯ÿÊgº:†¸jwü6¯Þâˆ_׃ßc™cËÁa}¸óW¼$ËðÈWÇï ë8äâ Ã:ƒûµ¹Œû_üÎX¨±×®ÞÛ~Ñ{ÿH_ˆ…èÔ¾ã4þÅýûÍPS8wÕÛv7ÝETaÛD¬èÑ»{;OD¨Jb_KBÜãwÛfµü6>ÃOaE¹ÁòÖ!ìiï‹iÚ1D»ëí2ƒ‚Ø<à 3(ˆÍ2ƒ]q伃*í4úŽ‡N÷U6alà¾<“èŒÚ1Zí;œ,ÕsÎãè5”Ä^=‚.íÕ#Ýhâê'l‰*n¯ôtP*"1X®Xtº;ç¯Ûst碻ôy¦ÿˆbæï.íþ¾s{˜¼tnR,ÇΧܤ˜Qe/ö”½ü‡×£‘;_ç󹃩;_Ò÷|±§p28¯Íáö`CXâ{ñŠâX„ÊT($Ÿ¢þTeËÜùO®íãÁNׇ?8ÝG°T¨Öëî'ir,†¥¼?=ˆe$ e2t{òt¦:–læ²³4£†šQ§ÍŠAäO#g9‚(áÑÉv4NduƒÊch;øתD„JD³yÄ5ùÑ÷7WÿëÐ0 endstream endobj 692 0 obj << /Type /Page /Contents 693 0 R /Resources 691 0 R /MediaBox [0 0 612 792] /Parent 642 0 R /Annots [ 687 0 R 688 0 R 689 0 R 690 0 R ] >> endobj 687 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 386.293 51.396 394.733] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 688 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [146.789 383.656 158.752 396.721] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 689 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 168.699 165.939 180.804] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 690 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [255.686 85.013 272.132 97.118] /A << /S /GoTo /D (noweb.NW2i2LRH-2zwz7b-1) >> >> endobj 694 0 obj << /D [692 0 R /XYZ 71 791.866 null] >> endobj 50 0 obj << /D [692 0 R /XYZ 72 754.004 null] >> endobj 54 0 obj << /D [692 0 R /XYZ 72 504.684 null] >> endobj 358 0 obj << /D [692 0 R /XYZ 81.963 387.379 null] >> endobj 695 0 obj << /D [692 0 R /XYZ 72 384.652 null] >> endobj 696 0 obj << /D [692 0 R /XYZ 83.504 387.379 null] >> endobj 697 0 obj << /D [692 0 R /XYZ 83.504 375.424 null] >> endobj 698 0 obj << /D [692 0 R /XYZ 83.504 363.469 null] >> endobj 699 0 obj << /D [692 0 R /XYZ 83.504 351.514 null] >> endobj 700 0 obj << /D [692 0 R /XYZ 83.504 339.558 null] >> endobj 701 0 obj << /D [692 0 R /XYZ 83.504 327.603 null] >> endobj 702 0 obj << /D [692 0 R /XYZ 83.504 315.648 null] >> endobj 703 0 obj << /D [692 0 R /XYZ 83.504 303.693 null] >> endobj 704 0 obj << /D [692 0 R /XYZ 83.504 291.738 null] >> endobj 705 0 obj << /D [692 0 R /XYZ 83.504 279.783 null] >> endobj 706 0 obj << /D [692 0 R /XYZ 83.504 267.827 null] >> endobj 707 0 obj << /D [692 0 R /XYZ 83.504 255.872 null] >> endobj 708 0 obj << /D [692 0 R /XYZ 83.504 243.917 null] >> endobj 709 0 obj << /D [692 0 R /XYZ 83.504 231.962 null] >> endobj 710 0 obj << /D [692 0 R /XYZ 83.504 220.007 null] >> endobj 711 0 obj << /D [692 0 R /XYZ 83.504 208.052 null] >> endobj 712 0 obj << /D [692 0 R /XYZ 83.504 196.096 null] >> endobj 713 0 obj << /D [692 0 R /XYZ 83.504 184.141 null] >> endobj 714 0 obj << /D [692 0 R /XYZ 83.504 172.186 null] >> endobj 715 0 obj << /D [692 0 R /XYZ 83.504 160.231 null] >> endobj 716 0 obj << /D [692 0 R /XYZ 83.504 148.276 null] >> endobj 717 0 obj << /D [692 0 R /XYZ 83.504 136.321 null] >> endobj 718 0 obj << /D [692 0 R /XYZ 83.504 124.365 null] >> endobj 719 0 obj << /D [692 0 R /XYZ 83.504 112.41 null] >> endobj 720 0 obj << /D [692 0 R /XYZ 83.504 100.455 null] >> endobj 721 0 obj << /D [692 0 R /XYZ 83.504 88.5 null] >> endobj 722 0 obj << /D [692 0 R /XYZ 83.504 76.545 null] >> endobj 691 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F30 574 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 736 0 obj << /Length 2878 /Filter /FlateDecode >> stream xÚí]ÛÆñÝ¿B/…yî‰Þïå"M‹°´y0ŠC^âÖ %úDX'©’Ή\ä¿wvvI-)Š¢(Þå\1æ(r9;;;ß³ ÝŽÈèí3âÿR€ö¿Ï£âêïÏ^Ý<{ùFꑉbjtói¤t,„iÁc"äèf:ú9bÉÕ?oþV"²¯ÒÑúvä.þñÖâ¦c“pxˆoO®Æ‚E) ¬‚.–£˜Ú€±¼¤£ŸÇJŠèyçÁ*Ê}-êè^ÿˆ0E¸ÂûóàÎagrV£¦‰¥5~Ž=3Æ”ÆF 54/ßh¼ÂMLTR°of‡¾|£h8"N´*ü˜®æé${~¥…å/ÓIt—mgË©»þ%ßÎÜÕÛô~³ÉÓ…û5Ï?gó|¶„q8PÀJØ h’ÄŠ²b ª>ºA:©€•d两N*–€¼ÂNzÈDÆä2FQÆÊGÑk„ïºâЦ¯ÕYF”¢×ïœXä‹OÙ:[L²K¤Œˆ–´”6K¼‡ô0S‘ÔXÁ¾Yéù á«Ç—> ôüdOºÎÓm¾\¤s'#¯Ò]¶Nª'±Òr/UƒÚ$0x;)Ï"q¾Tí'‡µrƒ[M¢i¯-\Õ-“çÑC¨;MÏ©¿µ/ 4iIL¹.<Ä^Ç)‰ 1…DÅDŠC1 †ˆ˜‰rÄ»u6Í'Vˆ.1?šÅ:IöΫ‘4´?åÄù{Âh.IcÆt1ì=‘„J€´³dqAc]óÿôÌ}$ k{™s ¶£ŠÁšáp¸ËÂ<)ÜqÝ#¼C˜¡ù™WPXÓBp}WcÉ£ÝÕX‰:æ:ú`‚PG0@ËèújL58x䣯¯åÜ"Z#\"›ct´c4ÞTç¿6˜m;*ñKÓ{‡Œ»GêýÂŒ¿øοZß ÏÄË”Ùm`5¬Yßú†eŠ·ë”ð‚…_f×—v§ŒÝR\dÛb`'Î!f× ¸S\ý¯oæxgQç:¥$ú3 `·T¬´#Î%ǹO†Òà ö¶¾ñf¿ñÄüSÅÆ+G±² ××W¸Ê:ŸÕ!Ÿ“‚Ïä4ŸQÖ9E汚ÐØd´ó…ãìÄó¾n°]Tà¸#}tÐþJ¦Xô‡!0Yl=Îê´°œmˆeßDþÒG­nUëÀöåTŸ(oàú¬bÄ@{‰¾ÀîÊwèv¤ k+Ì®×—?¹·½6>cw9¹»Þ"Ü]dc}ü&Úl,c1×¥ÿŸã.®œr[›?†XCy3Ú¼Cc2.}‰²¾$!… ‘QŒð_Á¦»;/a¦æ¥6ÃSMí}0 =í¸ÝÛKœãûKo}ë.³0w9 |q¡°‹…%ç‚Ëà]k*k'¬X)³”d¿»B›¼7²ž/„F²[j‡[RQMy¶jn‚(œMìºÚ ^³êÁr[‰^,Âg¼~‹0Ex÷7x Pó"*K €¶³C‡(Ñ…À†á-Ø7üa¬p2Dòµ Ï0 C±K¼×iKS¹é™ÓŠž‘¤6ŸõU¼Òön¶—%¢B9¥”Ì×¾phá40^|º î,}Iòìü¯Aôùº@ù0é¨ÇU!'9Âî'ßáäw8ùýºœïÝŸÿà“ß| Žv× v懵 îjVšh7—qd:Eí¦(9Ԇϗïq9Ûà Zc*£î¹¦Mø‰©:AJZU¥dqd›Ü=—<¾_¤ŽÌ“ ¸hooò ©Ü Ü5ÁnDŽ«˜E¼ ²¸Âí "JÏ/XQþ€rM›äZy¹fÝ}¨Á´5ƒæ&5±>0(AƒáÊÝ£ZÜ£Ô…/8*ëO² Dgˆ”£éPÒúdÅÞmlUøÓ*ÓRñ0eÚ®@²šTú&ÉËΗ2Tz%©S9b}¸Ä„r+¦4Æ°2A%u‰ÃÄÊŶA4àó&iý6”ªƒ*/2ÂÖõÁñ¸åÜÌlBUŠÊªRTÆߟîE)~ÍRD+,ƒÙß›|q;÷xf»U¶^¥ëô.Ûfk_m&!ÌðX ³¯#»ŽU\ Ä–…ëk´CdPP;_Þº‹å'÷×?iYæ?h³MÓtí—0;ä)æ–ÀcbÍlŠÌc ¤à¸¾&ziœ°\îG‰H¼&8HÅVòcŽ -š–¤`ø(ûA¦²M\ÆÜ[-“ÄÓ&bC@ !Ñë_ÓÉÖÕþ({I ³X§U껚þ5ð9a B«é2Û,ž_qŒÖíóuöïû|ퟦ‹»»Ye“ü=!lâ~Wd®³yºÍ¦îád9Í †e€4ÏÝ€Éònu¿Å¾ØÆJÝ$ HXÄÀÄ“í®]‚Ý!oL>îܪÝ^¶Ò\«ÿÚýØdAï„Ç´‘×äËÌs¹EñM÷âÞ §ù’#ak8Ù÷»ÞǪú]˜çyqIPðýóÇ6‰ü"“H„ÃåæSOÑïjyg“¸Ò$Òÿ:O'Б$t:5Ëä°FVmbcžüÄÏBŸc=»[IÏ‹ÒvO”¦æqµ5OG²aù2r¾ÅÆe‡F!KbÂåÃ4 Û…¾H©Ýù™Ãà²wT¨—lÉ|ÛqŠ)ïÓ_`ÜipêÝT5rw.Ȫ[£îk]iMvûì¹WÝ…_ùì³­SGûž7¦ ãâÛgúÀ &^f¶÷ŸVEݱ½ê>Ú†Âf3T4vÀ~@8E8C¸+ž–ßv¦“ÈAX—F›cíS`íS`Ð.ð 2ß× üAé°î!ZÚžæÿ†„ØkOëûÚ0ÈᾪIƒ’ÿ@jgügꡃ¦Ï-åÐFo‹ Ÿ!È0³G9ÿaa!{ƒWBy3—öëoÖ;žLžVT?Vî‚ñ6À§gwEsü‚E‹> âAÂqó¸Ý×O´1ªcN!å:.ŽµÕâÌ×7Ïþ gë§ endstream endobj 735 0 obj << /Type /Page /Contents 736 0 R /Resources 734 0 R /MediaBox [0 0 612 792] /Parent 642 0 R /Annots [ 724 0 R 725 0 R 726 0 R 727 0 R 728 0 R 729 0 R 730 0 R 731 0 R 732 0 R 733 0 R ] >> endobj 724 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [298.004 727.603 314.953 739.708] /A << /S /GoTo /D (noweb.NW2i2LRH-4GHEiM-1) >> >> endobj 725 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [277.142 703.693 293.589 715.798] /A << /S /GoTo /D (noweb.NW2i2LRH-1sNM4H-1) >> >> endobj 726 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [340.066 679.783 357.015 691.887] /A << /S /GoTo /D (noweb.NW2i2LRH-10PmPC-1) >> >> endobj 727 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [34.95 640.136 51.396 648.576] /A << /S /GoTo /D (noweb.NW2i2LRH-2zwz7b-1) >> >> endobj 728 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [248.036 637.499 264.482 650.564] /A << /S /GoTo /D (noweb.NW2i2LRH-2zwz7b-1) >> >> endobj 729 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 637.499 428.665 650.564] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 730 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [226.152 346.474 243.309 359.125] /A << /S /GoTo /D (subsection.3.1) >> >> endobj 731 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [34.448 293.2 51.396 301.998] /A << /S /GoTo /D (noweb.NW2i2LRH-4GHEiM-1) >> >> endobj 732 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [281.778 290.563 298.726 303.628] /A << /S /GoTo /D (noweb.NW2i2LRH-4GHEiM-1) >> >> endobj 733 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 290.563 428.665 303.628] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 737 0 obj << /D [735 0 R /XYZ 125.795 791.866 null] >> endobj 738 0 obj << /D [735 0 R /XYZ 83.504 755.001 null] >> endobj 739 0 obj << /D [735 0 R /XYZ 83.504 743.045 null] >> endobj 740 0 obj << /D [735 0 R /XYZ 83.504 731.09 null] >> endobj 741 0 obj << /D [735 0 R /XYZ 83.504 719.135 null] >> endobj 742 0 obj << /D [735 0 R /XYZ 83.504 707.18 null] >> endobj 743 0 obj << /D [735 0 R /XYZ 83.504 695.225 null] >> endobj 744 0 obj << /D [735 0 R /XYZ 83.504 683.27 null] >> endobj 745 0 obj << /D [735 0 R /XYZ 83.504 671.314 null] >> endobj 723 0 obj << /D [735 0 R /XYZ 81.963 641.222 null] >> endobj 746 0 obj << /D [735 0 R /XYZ 72 638.495 null] >> endobj 747 0 obj << /D [735 0 R /XYZ 83.504 641.222 null] >> endobj 748 0 obj << /D [735 0 R /XYZ 83.504 629.267 null] >> endobj 749 0 obj << /D [735 0 R /XYZ 83.504 617.312 null] >> endobj 750 0 obj << /D [735 0 R /XYZ 83.504 605.357 null] >> endobj 751 0 obj << /D [735 0 R /XYZ 83.504 593.401 null] >> endobj 752 0 obj << /D [735 0 R /XYZ 83.504 581.446 null] >> endobj 753 0 obj << /D [735 0 R /XYZ 83.504 569.491 null] >> endobj 754 0 obj << /D [735 0 R /XYZ 83.504 557.536 null] >> endobj 755 0 obj << /D [735 0 R /XYZ 83.504 545.581 null] >> endobj 756 0 obj << /D [735 0 R /XYZ 83.504 533.626 null] >> endobj 757 0 obj << /D [735 0 R /XYZ 83.504 521.67 null] >> endobj 758 0 obj << /D [735 0 R /XYZ 83.504 509.715 null] >> endobj 759 0 obj << /D [735 0 R /XYZ 83.504 497.76 null] >> endobj 760 0 obj << /D [735 0 R /XYZ 83.504 485.805 null] >> endobj 761 0 obj << /D [735 0 R /XYZ 83.504 473.85 null] >> endobj 58 0 obj << /D [735 0 R /XYZ 72 407.54 null] >> endobj 62 0 obj << /D [735 0 R /XYZ 72 331.8 null] >> endobj 762 0 obj << /D [735 0 R /XYZ 81.963 294.286 null] >> endobj 763 0 obj << /D [735 0 R /XYZ 72 291.559 null] >> endobj 764 0 obj << /D [735 0 R /XYZ 83.504 294.286 null] >> endobj 765 0 obj << /D [735 0 R /XYZ 83.504 282.331 null] >> endobj 766 0 obj << /D [735 0 R /XYZ 83.504 270.376 null] >> endobj 767 0 obj << /D [735 0 R /XYZ 83.504 258.421 null] >> endobj 768 0 obj << /D [735 0 R /XYZ 83.504 246.466 null] >> endobj 769 0 obj << /D [735 0 R /XYZ 83.504 234.511 null] >> endobj 770 0 obj << /D [735 0 R /XYZ 83.504 222.555 null] >> endobj 771 0 obj << /D [735 0 R /XYZ 83.504 210.6 null] >> endobj 772 0 obj << /D [735 0 R /XYZ 83.504 198.645 null] >> endobj 773 0 obj << /D [735 0 R /XYZ 83.504 186.69 null] >> endobj 774 0 obj << /D [735 0 R /XYZ 83.504 174.735 null] >> endobj 775 0 obj << /D [735 0 R /XYZ 83.504 162.78 null] >> endobj 776 0 obj << /D [735 0 R /XYZ 83.504 150.824 null] >> endobj 777 0 obj << /D [735 0 R /XYZ 83.504 138.869 null] >> endobj 778 0 obj << /D [735 0 R /XYZ 83.504 126.914 null] >> endobj 779 0 obj << /D [735 0 R /XYZ 83.504 114.959 null] >> endobj 780 0 obj << /D [735 0 R /XYZ 83.504 103.004 null] >> endobj 781 0 obj << /D [735 0 R /XYZ 83.504 91.049 null] >> endobj 782 0 obj << /D [735 0 R /XYZ 83.504 79.093 null] >> endobj 734 0 obj << /Font << /F57 146 0 R /F55 149 0 R /F73 223 0 R /F61 150 0 R /F70 201 0 R /F74 202 0 R /F59 148 0 R >> /ProcSet [ /PDF /Text ] >> endobj 793 0 obj << /Length 2669 /Filter /FlateDecode >> stream xÚí[[o7~ϯÐË£$3æý²®÷¡@lûТ0úÐvH‘b Q$¯%Ûu‹ô·/yÈ™á\$qdºk z4âœ9<äùÎUht5B£·/ÿÿõ勳o¸aTh¤ñèòÃH’‘DªÀ\Œ.g£_2Vð‚ŽsŒÊÞüv=¿›ì–›õ8'Re?Ül®'WîÆ.¿­è–ÿ q9R…‚YÚ9âl”THEÜ °œØgϾ‘4äÃOEnÒÂM8œÂ ªo~p,-×æ7óõû¹ûx¿Ü-ÜÕÛÉív»œxÎWËóÕr±ÙÌåŸXÑB(Q’ÞÃ!f«9\þŠî£ÅPA‰.§ýŠ8ÂÜŒ0÷6÷íßÇQyõš6ˆpBc¸†BŽq¡9÷<µä ›«‘»ø$Ï"†m%*&–ãœòìC‹DGsDÜ¥!3ú%œdFhTg“qÎPv×W0.áNsïÍ#³ì+x‡ð_ÜÒt´áòÇ·á›{¾5dˆ@Œdÿh³y )a˜]G›$äÊffí*›Ãx£•†Ìîàzã®sžæõ2ûdȳ;Î`l’Ž9YþT4iPi#B j”À£ôç1H÷Âüã"Ëz 8å†'–=˜»ØÞÄrl¦ßŽs…3P€œ‰¬0#ÏÞÁ5©îˆì FGG‰l Çm ÇØ7e¯€æhš;’yšfÚY=˜wû¸çtSëÐ JdÎuë,Kí¹0SŒ/S”<»®–RÈ3 (qJ¢”JcãØpÇÉqo¤§ÂÔ(_Y&”à!ñº‚Å|맱6>àÍJV“•«}=y˜ocëWÜø7­úÕtOýÊN®_ý46èØÏY²ºá¤àJÕu­éžº–®9ßW×’æÂÿÖe­4%–S ±ëÊ“º\™Ì‡ #×Ї³wtÃ{ÛÀ¸毒0aLë ò ò}ÃP'q©÷¡—¾¶ÌŸ¶¤§À©ãýÆ5¬fvJ=ÁîðzÂ3=FÆðÄGüà[Æ:Ûè™,%Èl_¤rgæ€1¿ÞDº3{ïÛ­"Öx1ë:ÅfÆòÔ›PÅfáï‚êeYÉŒ"¢h+žˆ{usœE>"=(üw`ÙÑ2Iåb†ªrñq>ûÊƇ¶ée´oÛÌæÆ>ÕÑ›ÁH%úë†Ï©˜ÍMZ-@XóTyá~#ÀÒ%ž³Mcûň‚k¨¸—Y§UŸ"£Êþ™ÊM£Áíe´áÚ¦êQÊÓØ) ̤'fª›‘ÂPP306‰†±dæ9¥üC¼ ½ÖA²©L—ÑÁ³hìŽÒÄûŽÓ»HUm™FkÍE´Œ¢ižEwômÅ©®8oš+_[5ÍQU§O [ç.[³7ÜÍ}A¬ºœe¡ö.¤UOû™wës”q…{µ²ËràÓF.]'uQ°‚ŠÛ‡’Ûú“ä%TM6¾êV×á¶Pe©úcÖ0í5|…Ë&wq~ZOo{ÀÁ™ÃÓÆÎäA÷N-ƒP~Ò·î¼ FИåjHÈ7 Në..ßH²1dCiµÜ"NÈC‘GàV-ÊspäÊì÷Vݹ;íè Œ`=¼åF~ÜoÈ¢Þ‰Áƒs'ÍYàf/Ã'NWcà„¥êˆâÈçmWÁèZ ßYÝÜIäxj»ÂášÃ+9ì‚Ôˆ×`pêoûL””!žGÇûÆÛ€÷uš×ðF_é•Ç¥ö@Å£ÎÒw¡„¨S-£Ii`ß ƒ¡´ö|¼ûhÕ÷·ƒ6H¨ñ¹BÀ5wrøVùÆÑÇÔÞ¡8]Wþ÷Àéw©â–iÃÖPtŸêȬƒpΕ ·þ•‰J×w­:ð­È“N£ò©zÕ²¢ðó²¢'ÈŠ‹AM4³Ô5À9]ÖÌ ÎPõ{×(Põ©iŒBU ÜU¨J‹‡žv•¡ |£T$,Æ>íz_uEä1?ÿÑ&|L–<]û\aìÏ*%k¹+³ ‘nhÊ×sì;{o‚®·°Çí\Ï ¾;©Ç²ƒ:§v˜°'ê01Øà4á5\OËkÁšÑÚEÕ{ GŸH;ÇßN, A4µ¸\#èÆäíHÚXß“zTØã{T¤ŠíÃŒn1T-®x<:>ëdÒIæ´t7Ä1!höY±ó#8Œ¹KuJþ¢ì7Øî]õ±””ÿÛŒA"É¿¥ÜP|UíPA´Wþ^V»"R5…V]ú_ö„Ûþ)þc¬zGóÔ3³êÄ2ݰƧ·t ôt=Ý.YC¿ôtÿÝzº™K©è{º;>£þkúšÈAs.Á² ¶Ñœ UhÆzmõ›ËÿPºS endstream endobj 792 0 obj << /Type /Page /Contents 793 0 R /Resources 791 0 R /MediaBox [0 0 612 792] /Parent 642 0 R /Annots [ 785 0 R 786 0 R 787 0 R 788 0 R 789 0 R 790 0 R ] >> endobj 785 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [34.95 686.288 51.396 694.728] /A << /S /GoTo /D (noweb.NW2i2LRH-1sNM4H-1) >> >> endobj 786 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [258.934 683.651 275.38 696.716] /A << /S /GoTo /D (noweb.NW2i2LRH-1sNM4H-1) >> >> endobj 787 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 683.651 428.665 696.716] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 788 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [34.448 464.634 51.396 473.432] /A << /S /GoTo /D (noweb.NW2i2LRH-10PmPC-1) >> >> endobj 789 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [327.836 461.997 344.784 475.062] /A << /S /GoTo /D (noweb.NW2i2LRH-10PmPC-1) >> >> endobj 790 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [416.702 461.997 428.665 475.062] /A << /S /GoTo /D (noweb.NW2i2LRH-FUd2L-1) >> >> endobj 794 0 obj << /D [792 0 R /XYZ 71 791.866 null] >> endobj 66 0 obj << /D [792 0 R /XYZ 72 727.375 null] >> endobj 783 0 obj << /D [792 0 R /XYZ 81.963 687.374 null] >> endobj 795 0 obj << /D [792 0 R /XYZ 72 684.647 null] >> endobj 796 0 obj << /D [792 0 R /XYZ 83.504 687.374 null] >> endobj 797 0 obj << /D [792 0 R /XYZ 83.504 675.419 null] >> endobj 798 0 obj << /D [792 0 R /XYZ 83.504 663.464 null] >> endobj 799 0 obj << /D [792 0 R /XYZ 83.504 651.509 null] >> endobj 800 0 obj << /D [792 0 R /XYZ 83.504 639.554 null] >> endobj 801 0 obj << /D [792 0 R /XYZ 83.504 627.598 null] >> endobj 802 0 obj << /D [792 0 R /XYZ 83.504 615.643 null] >> endobj 803 0 obj << /D [792 0 R /XYZ 83.504 603.688 null] >> endobj 804 0 obj << /D [792 0 R /XYZ 83.504 591.733 null] >> endobj 805 0 obj << /D [792 0 R /XYZ 83.504 579.778 null] >> endobj 806 0 obj << /D [792 0 R /XYZ 83.504 567.823 null] >> endobj 807 0 obj << /D [792 0 R /XYZ 83.504 555.867 null] >> endobj 808 0 obj << /D [792 0 R /XYZ 83.504 543.912 null] >> endobj 809 0 obj << /D [792 0 R /XYZ 83.504 531.957 null] >> endobj 70 0 obj << /D [792 0 R /XYZ 72 505.721 null] >> endobj 784 0 obj << /D [792 0 R /XYZ 81.963 465.72 null] >> endobj 810 0 obj << /D [792 0 R /XYZ 72 462.993 null] >> endobj 811 0 obj << /D [792 0 R /XYZ 83.504 465.72 null] >> endobj 812 0 obj << /D [792 0 R /XYZ 83.504 453.765 null] >> endobj 813 0 obj << /D [792 0 R /XYZ 83.504 441.81 null] >> endobj 814 0 obj << /D [792 0 R /XYZ 83.504 429.855 null] >> endobj 815 0 obj << /D [792 0 R /XYZ 83.504 417.9 null] >> endobj 816 0 obj << /D [792 0 R /XYZ 83.504 405.944 null] >> endobj 817 0 obj << /D [792 0 R /XYZ 83.504 393.989 null] >> endobj 818 0 obj << /D [792 0 R /XYZ 83.504 382.034 null] >> endobj 819 0 obj << /D [792 0 R /XYZ 83.504 370.079 null] >> endobj 820 0 obj << /D [792 0 R /XYZ 83.504 358.124 null] >> endobj 821 0 obj << /D [792 0 R /XYZ 83.504 346.169 null] >> endobj 822 0 obj << /D [792 0 R /XYZ 83.504 334.213 null] >> endobj 823 0 obj << /D [792 0 R /XYZ 83.504 322.258 null] >> endobj 824 0 obj << /D [792 0 R /XYZ 83.504 310.303 null] >> endobj 825 0 obj << /D [792 0 R /XYZ 83.504 298.348 null] >> endobj 826 0 obj << /D [792 0 R /XYZ 83.504 286.393 null] >> endobj 827 0 obj << /D [792 0 R /XYZ 83.504 274.438 null] >> endobj 828 0 obj << /D [792 0 R /XYZ 83.504 262.482 null] >> endobj 829 0 obj << /D [792 0 R /XYZ 83.504 250.527 null] >> endobj 791 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 832 0 obj << /Length 3162 /Filter /FlateDecode >> stream xÚå\Yo¹~ׯè·Hˆ‡!Y<,o°Y`±r°ÇÃÈÛãèZid¯ÿ}>ö5M›ÃÑHöû`4fÉ:¾úªÈ6oÞ6¼ùö„÷Ÿ_¿:ùÓß´o„`^kÙ¼zÓXÙXEŒ+ݼºh~®¯ßv¿^¬îÖ äê¼¼ë^¯V«‹?Ù Å$™fAœq®:‘—ý¼–Oç ? ‹éÛ1·ýh§Xdõ#~âšçä`B¥†AŸºŽ'bÜ0à}N„d$iñ&·Å´““•@á^¨Ó¯Â.rzUÓŽÏŒ:ˆÄzƽÛÝ`"W'së7Œ›QOç3š”:^¿”ãÂcu Ç4éÆ2rzÒTQi.0.k1ãÆŠÃŒwòÍ«“_NÂC×tؼmŒ‘ÌCM¯¯N~ü™7øÇïÞúÏÇvèUƒ‘Ž%\6ÿ>ùgžñrˆ<“P¾Ñ’­'úlBr&±VüÀ„éƒå/9{Yæ¥Ý±û>ß·%aïLcHc¤,nK3éJ»’oéDc0ú•g-&=P¢…Ω÷Un‡Â3Ë1Ý°ûõÛ¬ƒcÖÂOéJ;Aà¢ØU»U ;¯ÐºL)QT»‡~ÈÆjïÉQ°ÙBÓÞ¤®ÿ±¼_°à¾!6Œ‘9cf•àŽ£írŠñàüæ‡ÄÂæî ‹7;’°X»k4” „ÙoþàòŽ¬ëEj¼D]šb°ô‰ó rLVE>žèkÇÇk,DÏj¡Þ;ŸÄ@Ö2Pø²¢¬(‰Å›½V¹À§žÈ@®Ò@¦` ’‚YøV……Œ¢¹’…aÕô¨¬4kÌ£"…‚Ý f¨_Ñui=&ß þ0·tT¶Ÿ-úüH컾•$8\?.*»ì…p†9éÁžAô„~JÕgÑY0‡iÄ­Q¦üa=TLMƒ0ƒèAqñ½Æ¼dA -ÂD2²ÛÌäô•œBCO{È®°Ú~v*áp\øØjY¶ä-ƒi\¶B‰·'rµæB*P! :˜e€ÇPrʾäüæ×ÛÕëMW]¶åà?înn—okËÍÇo'·_Ü­~yXß Uçúz³z{7ùgŒ]_ßoú_ºo—÷÷ëe?âjµ¼¸«Ú»þÛ›«ÕuÿŠã×ïPé²–…ò¶ÞkWøŸ3«Nñ4yjKÝð“:ÝÜt߬>,/–›U÷íL„€ç‘]퇙¤7âw’ cËo‡‰°ÈÝP„'™ÞîÿÊùlÈü $t;ìû||¸/T{§õï¤ö~‘_†Ób"`v™t„Q°»Ù–}óy µ¡×i*ró„éï9I¨±½­Qoª›t¢‰nö—é?qiæÐs©8é£ø2·£¢d>çYsM]t8¸Õ·¼ïöãêò² Öá›Í»>z§-®]÷míÃʼ+NgvæTNøIGX§ e L†y{Úr@idÖÆ¢-ƒ9êy·Ù±ñ‹.K’ƒS¥pržU·fœŽ€…úý:KõzfKfÎgZ„~ÛÍù*ÇßÃÉ4²3ˆ2½ª)u ¨·Løi¼%,ݶèQ]Jç £¨2mPL±Ÿ6„…XE6˜ÁZÙDŽ®B6 Ýìv;[h§O_†ÚI· Òšäòî~Ón)ð\í‚S¾ïóÜoV·/ºnËÇÕÐعxn[…KŒ(†È"¢¸:.] ²°ÄÈg¸…FØIW @•‘Å9a(k¼–‡P¬³`xÏÈš‚Â\[:£äD5cŽ$)ƒ°ƒW/ò‰Ô@¾$æüŠDÚ£¯F|&=ì’ZÚFôÝU°A£Å~€®éL°î¼¤ŒUqýi•i€ $&©ˆBËû>E‘¤eàj{{ú`Òx0 §©Rv<@î_{fEgÙÔw–Íq<@‰p·‰É@9ZÁ£,­·Á-ó©w(ïóu)q߶Õl}Ù=¼H¦Œ ÐÑŽDÿ"ãD"Ö›QiQ½j“[ušq¡K €©8*:*Ü\&b†«é 6‡ÀØL–DÕéE11+ìĘâ ÃA>M¸ÏÌúàE2{˜v/ ÆE_P˜Q´aI™.aÓm—.]]º”}º¤št^Ün¢òîµôõðU×þÞ´I¯4k…M´çóY£;¼‚Héç8ŒŒ·ˆ‚|BÎ¥™¨ò@]¬Îøgȹvz8°6vÚ•¯9ÞLc>œ9„7¯[–9`¸Œ© }yR¾ê!*4[åÞ¾¼Íõå7Dƒ%„7u”ÊÀètŸ± ͛”–…F¡†èn×ÏÄ…·-ñ}þ~ˆP Ûi·3ŽÈ¿f"òeöŠ:¢ ªš„dýiBºr)}õ™U6:ö'×=§5=q¶Æ—Óó—äŽÌ|ßš¢¾õ$ùš§‰ù…Drá2axƒSˆB_[æã*ûÚÒÈÊånx‡ÏÖ$Ui\{=ñuKeôóœi}1·ú (¥.,t8a£óîì´j<"®•çOÁ9ÂÍ‘ªiBÇ)žæ˜ªP±?Û Ur[V„›&”@"w¯ÍÖ¼yTDœO·GÄÑíÔá^D BÉ–¯Nê‡};¯â¸Âû&¶îx]~N(’O E¢ŠÔ—º±=EºŠt‡úQP¤+§±N%Ó¡H×A‘~ (2]d‚V> endobj 833 0 obj << /D [831 0 R /XYZ 125.795 791.866 null] >> endobj 74 0 obj << /D [831 0 R /XYZ 72 754.004 null] >> endobj 78 0 obj << /D [831 0 R /XYZ 72 733.159 null] >> endobj 82 0 obj << /D [831 0 R /XYZ 72 550.051 null] >> endobj 830 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F70 201 0 R /F79 203 0 R /F80 204 0 R /F74 202 0 R /F82 566 0 R /F77 565 0 R /F81 206 0 R /F83 207 0 R /F85 686 0 R /F73 223 0 R >> /ProcSet [ /PDF /Text ] >> endobj 836 0 obj << /Length 4939 /Filter /FlateDecode >> stream xÚí]Ks·¾óWl’‹UÔL4žW9VÅNùª$:%Ê‘(‹‰DÙÅNòßó5æ±3Ìcwf—”ƒŠÔ `Ý_Ýh`ÄîÛØ}}!êŸ_>½øõWÆí¤(ƒr÷ôåΩÓT mvO_ìþòÙÇ›»Wþúô›²¥f'Jémõ§gŠÿé×_9Ñí¢¥7zWt› e릡Û4”ŠªÑçu ½sep"6ªÔŽvº Ö¤]yê4´˜rhz*2=Ù2øv¨¸ÁÅoŸ^|!ñ¦¿“Ò—:X<½)=ºzþæâ/»øã7xjMv÷16}³CKéˆÝëÝŸ.þPI1™µ*•Ç'ðS5'jçeÐÖYn‹ßȱÌd¨…u[µê¯‹¤’¬gA˜P5»º}Q- D$ÔÂ…q>]˜vRÉÂX’“ cJ )éR…Ðï*«×%I¹H¬MIµéj;©ÚÒèžPËGióÙÓWÞ?~T(ç?ûx]ý|þööùõíÝ»«»úƒ··ÕÏ»WõïoÞ|÷úú]õŸï?\ÝÞÝÜýPýïõmµTdK¯ÜN•Nº™µB¢ø£iúçܸRá‰E3{¥ [µvWÀM-§×·¹QøWÙ~ýÏ9MB¢ ‰núRµÊÒ „$!j¥'ÇÃÜ‹n»ë}W÷(“A] ÖCm| *÷¾P¬¢Û4"´\kÚ›ƒÉ®Ô ë'|™ÕBØ)),†vj–¨™P[:Ý¢f/Û‘÷vSMžÊÆTÆJ•Z†ª$5i¥„%"Û7ÓTt)ÈìlPeÕ”——Qz‘Z¨2©¥§6ÜÊý™²Y€®È%>£ç!(QR wJþ/·ún()EÐ¥€HCS’‚KJ½DæM_6¨†é Ê.úÏ(lû”Ï Œ×-ÕPŸè°Ù;À¿çLŠûЩEee°Ž®oQ/r³jDг›.¦¨eÐ;U7y’ÃŽB»Ò QaK¥jœþ1³2ÎE—äà*t+ÃÐð*äw²˜é»ëÝËf-úƒt„¼k¿Gùï†ÃÚÁÔÆ2æèþ0ÿÎél¼O–ÎÉ)`]í¨] ÐÆ­ç”Ú¤mðÜx‰mðôD˜2Ž¦³œq¸qãÀäežÐë¦ñ¹²þÄ ‚ÝØ´ŒŽ>{ „hú:\”¥À–×ÛRŸ%z§,Oyи!Aû}™óûpÖuO¿Ì™¸¹Å€oÀuzßn%\ý"àHb¡di¬ê­7ƉäˆB(Ï‹í¤9#1?s*À¤ßY‹(˯¤"P¨»ÞYãK+ 1‘ŠÔüör”ŠHƹˆ1p:jg<£\ÉEš¾;o—"ÍËœ=TLãäÉ2-² öÀÜÂ@ÐÍ­dZLÚTOG&Êú#Û|Ä™/“ùÌ> MªMçP•óOŽsäÌ4uÓ‡Q7ZMÝÜrÔíJM ¥öeFj_f¤öeFj_ä£þì×ô»¾ ÒS/³hhðH Ñœ*ù7­Æ# VÍTŸ{aFë]ºa†‚ðmà®–Á›v±÷Z¶SehMºÔ½pbS:£ÇcÒ1j9Á ¾WåÀ# w2”+íª(«þöh쯌+ g¶ŽÆ ~úþ0ÿÎ-DAÐ\ccìoÙn1sÛ-Ò¦$iàÔµ1[Ðâ~Ø -IøJ}ì ¬Zèq¶kõ@AK‘ëØnÓ—$5P3ªðfHHVÑ©ËÒ©Dò̦fÂ-QÈòö——Óû3–ò“rÕˆ· O+ÀP‚¾—x«ð†JÔ ¨Q@9ø'€’ø'€ÿÍí'«»–¸Fo×ІqÍÃõ°j‰‡UÏÃú%Öö<ì帇•s.V#Pr°þã|–¾'Ì»Xmñ%ïŽÆéR«pkw±Å¨‹µ÷âb!™i«"8|m Ûöõ³÷±8WùX­á[ºÿ÷±ÇûØ\àßõP4œ¾ËLße¦ï2Ó7#µOݹE˜T¾G‡m÷kdoÃ&ÕŠèZ:ÍŽÚIœ±R09·ßî|“±7˜RÈxã^"!Jßšn\™°^Kø;ÐÐzØOZ¤‹hSõV—‚ÃãKj6ÏœL äÍsR&h™fÏú³Îž „µzÑΛû ＤhÊ ßd¤XéwܦuýŽ’¡4W´ŽÏ•+ƒÕ7KLŽÇÄMô‰¸HHê™”‘eÞŽÏ$ÛµL¤lƒzvÒ2ììnN&lÆJ0Ém “·Vˆ•o­¾èûNŸƒÎ±°Û}ºÛ‰öÛ‰d”sÈH=d¼\‰Œ3¥:ÂËm Q›9ÓRõvsh\¼É&î²MÖ¾YˆÂC‚¦Á½³DtñqÐËø#VBè)Íh;»G§™š‰iËeNÀ­¼× ÈRÿN N‘k7 N`í¬©Yå²Ë•ËnéaÕâàÄÜgY ÓÁÉïófB.;ë8競'*¿güæTnºõ›u£ÇQI° tžýÀ0LVf=7>& 7Y¦Šƒ _7ã|]Qéá#7R:ªÛj8ѧ‡“š¡WYèú/:˪áIJ!ó&­}zÆpiáÚr˜õåÎ Yr²ôauydé-HÓ¬9ý‹_ßÓêtLÚ®fÒB€Ÿ¸Eè4Ϥ›Îî†LÚg 5×=™(æ4­‚4C9}Œovû®íJñ–¦:ËÑ ækŸh7GAÖF }o¬s½™XbÖmùÉ ƒ®”ŠÖ Tj<£ö¼…nxÀÊ|¦x&¤ÓhšT=Ë> –PxvpÖ>É´K1v“uRÜóÒìUÉ –†â9I­\FèPuƾ#Ââ\¹îSXÖeîL©t«V 1"ý,Ë,¶¬p°eÄ8\y“u"Xù°,ëÓÜEšˆJ=3½ÑŠQ…õ–µiT–…ÙRY½nÁ½ºB'ñR%)àmˆ—*9eïëV¥lˆ«ö ÿ*ßòÅ(E·Ý YP=·!òU»&ú¬NÓì)T<†vs'Ù¹ì5˜f¼PzÝÙž¦/¾¦-¸¾ juõo_`•÷Lä•-`ö (²/¯%Á©Ù&8Íl¥e`*r|­à[W:VÄÛV´”@ªp mÃrm¶bÐùå ŒQ¨xqô†zÈàÚÞöõSÕöæ¨<挜œ5o2$kåcš-Ì‚¼ˆg^Ïd7Y¿£ãÑü†qØAÐÍvQøõvQN›1×VÆPm [lûØâ*6î¦RæD¾®0ûs +ù E9·äôn€ìÝ”1´_»ÃNLþ¥Åj|£3ì–6š¸.×Î/ÂZ‚L‚“Øplâs;š#Å‹¶Å(¹&ÈżÇô…“ãÓ9€y½ò’ º+/÷;ɉ¯±+ö“s‰ÎÅTÖÛ¹:FËɦ„¤ÉéXÆXŸ¤ª®xöH>¬‡Q«ùvu^0wÁ€†™‰£‡òZô†ùoÆæ9¸àšL„’mUñUæh'~ãP¤s½íev“XabÌ3€ÐrÉÁ¬ôÀUâ<$&Ð5¬Õ”jK“7[šüQÙˆƒ/3øbÌàõÌýƒ|W †ÞÂàÛ¾Îcñ6ÀLܜœ†Þ“Ú"§¥@9—ÜÀÉí³&Oz(ŽF E=oò€ ¸Ãq°É9Ã"ô 2x¬É×Γ/{÷‹ÎBΟqœÅ‹=Ó\¹ù5dšvÓ´gIûZB¹‰Y¶} Ìòtµ|Y9WÕf4M—ÍÄPó:wy0Gw#Š»q¹KÒ|eàÁ#ÎÇ•sS2 ¦Ô˜²™p r¦îœd0‘{ˆ›3g¯Õ­¯¾X²É B6å@Úðw2–×v¡­»Ò6åÂ+ë‹Ø’ýÅÜßÃB–¾¹¿ãã«ëw×¹qIÆôÎúÈ¿)»TãóÌzK çÈ·DHá:¤ûÅ©¼R¬Bšún=)N"ÒÇDOtz¹ôðNèÎä=ªÆwu;©çæçùR“ÞüFßu!ùŒ¬³´›:äJmj(ðw¹Ú.DAÐ<âúÈ•%}m_|]é “f|¶ä½,´“] ?óïe ¦´p<¬@ûôµ6V õ¢ß%`OÓü{P€Úñž™%1󔶫‘§H߃âsïA¤­µ‹ÆOãik¥1uŽb7PŶ¯ž*>É­H’„Z÷…€ÊƒH¹*g«ÜÉ°x@HJBFÛ¨]Ûק„€ÄÇ…ÛIL[‹€Å ă(†-°íêÌH÷å6ªØôÕWÅXuмڋÉSæŸê;×´ˆciÍF(XZ/ëÄ’w®‰¨ ׃=Þ°Lmøµ-Þ·¦Gj +k»ZceÃ÷­¹þûÖbq&RUpw„/sWÌrÆK/¾â};É૾¢ô ºðäâ«¥¦o҄䥟â–:{M¿@­ö07ªŠÝ$§ìÖk`ÓÕ" TC Lœ•éhàâГišÞb(tÌ©õª!“U6.œd•Óë§)f¥¶Ð⦯¥ÅÀføªT‹§@-·O깸Ûn¢æuWǪyºCèxûv¯æM…çÝÕó¼®&„z÷þ®®½~wóϪŠ4çÚƒ—tZÇú‹ñ½IÞ(óƒ´ç\ÁhºÉÍA°ŠwµY=yJ–·nT¢{ý’lØÑ:„àÕ¦t5ó1¶H<`ïU#ŸgŸa…0Ìb„œKØïLçëíxÖ…5{.OnhÕ¹®¸ÎJÈagÃ÷Z>&¡ôÜýŠ¦H(oôóK„•!Àj˜‘­ä›·\Œ¯ôâ{v6“k±¥\‹ü‹üârG²ðI^®i!+B…XÅ4k+žW~ÚXœEý=cßø¡R6p~PÙFv3 1¦q›èÅܤòŠ‘„2|yrñ&Õô×Q^ü茚+¬d,#ÇôÈ|Û9Ñä*IÞå>±š1;&­³€†¨œåÓ)!ýbBÕæBñFN>DÓÙÈC$±©‰'W:1áZÜ'^åæã PÒ(ÕxrF£[¾ê“(ÚϪäLÓ—‚wï]~ŒI¾$FÌ `¢N(!–2ô`Õž½¶$>a«øMÌmqhd[^äÓÝF ïhý²H÷ .‚îâ=óÒcMä$îkS¦g[lúr>^ÇÞ.Řauí½[P›êÃBÞ¢xoÀÍZŸÔä·{ ž¶ZgQu_2¨¤6:e ü²ç%‹¸À ¿ÓYWñXƒâÈxÒþ&Ã*d׺z;GÇ”¾Ï¦PùØúÜÞ·õ~‘Pø¶¿I7CŽ·@ô¨PBr´€“ÀÛ¨“f4Šm¡NM_³êÄoѶv#} \kïVéSBª‚å"|ò›š÷;/%UÅ(©’†¯­0Û*©Q¿”TÑ2RegIÕ‚‡XNªÆb”TÑ HÕl‘Û눗Q*f$¥C¨aåJP÷ee<8N©ä&-ÿJ%bUoùW·aà]õ>£ÊP)¾ûÁtU}”I¹MjÍÃ"áf‰Tc3ŠAiòݹ6Vñ/±?¬Ä~ ‘Zk5ǼE!Ò(êA±‚F)«©±F19´®€¡îKU¯µ"LAϯá#âÇÄš ÖðX#ÊШ®ìz×Ï/#P¹T”Vq‹f4æù ÄòÍú#ÒãOè‘—ña7Ñ£¦¯9=rüÊ,½9Šl%ô ³üå}r.¥OýþτÈeLd7ð-xs›<ÖÏ( àö|™¡ZÀŸl6-E£µõx—?ýl=e"ç& gD >Hм@`:JÙédjõÆ®@ÆêÂ؉ºòh¦ żµp„ ¹ONò®´–º/vþNŽ¿I‡ƒ cˆjÖVß4"ªƒoíñ #¶ƒ¸±îâBì¾Å$¾¾ˆçÁ)ðYËê\Sµ•zð¨ÿ”ªçá endstream endobj 835 0 obj << /Type /Page /Contents 836 0 R /Resources 834 0 R /MediaBox [0 0 612 792] /Parent 838 0 R >> endobj 837 0 obj << /D [835 0 R /XYZ 71 791.866 null] >> endobj 834 0 obj << /Font << /F57 146 0 R /F70 201 0 R /F79 203 0 R /F74 202 0 R /F83 207 0 R /F82 566 0 R /F81 206 0 R /F73 223 0 R /F80 204 0 R /F85 686 0 R /F77 565 0 R >> /ProcSet [ /PDF /Text ] >> endobj 841 0 obj << /Length 3510 /Filter /FlateDecode >> stream xÚí\M·½ëW´“‹ch²øÀ€a àƒ BÄöa-m¬ä•íUl$¿>¯z¦gš6ÉÙîÕ®€\¤Á —MV½zõÁbËáÇAy&ÿùòÙg_Y?()¢Œjxù¯ÁÓàÒØáåëáÛçWw/väÃó߯߾Ýš¾yÿæzÿáîúÕ»Û×ûϯ¯½ùíêýÍo×/¾ùõg_y9Ÿ\É(b0ÃŽ‚pÆíŸðÅ~$–áEôrè†ZÄÃ:Ž™Íæ„7aØF½½-?TDå°ßqÌ?yȳ?¿|öË3…ï0rPA«òX]PaxõÓ³o¿—Ãküøõ …Ñnø}úÓ $ññíð·gÝË/{ZÀÖT\ŒÂØâw’\a›ŠD° ûTV'õq>'‘ðDƒÚ›ý°ÏÃÌl2Ò‘”…hÍL´éfÉᣆž•PN×6«”°Ae›Mž8Í¥Œ°“f{ÅÏ%à ¯/B„ãX •%•9a•v ‡i¯ ;¶ÂX3CÉÕ᱿DïÝ~Lô˜I¨xô§…‰¼0:}QšŠåñg“eŠ¥.{AÚÖ¼°ÀB§€¡yy&`UBTQûTÀ%¹à±`Ÿˆ( àE } Dçú˜µÛQ»‚€¶ÜQÚ5½^¨œU4¦ó=ªŒœ¨Æ¡øYH½çЉkÚ:²çE„³ã°hC:,÷%Ϋɗ,­Év¬ÉÅ"†’EÙIŒ¶ä’a™–xQÄZ2Óâ¡%*­? "¸ß ¤×¹÷"’ŸvYñª¾âU lÍHÛ ‚»¯¨¿›àa³iàøÌhè( ]ð¾A ™:ßäÙ;- ØÆÿAO‹4¨Œ>´ÞÁO×M`ïökQÄ4Ìuüªe¦{ȶ\ÜÍaÏ#Vb Å ™¡÷EËrÓC¥+=N 5KÉÐgÁ·ÌÑÂélÂ<^3ãI Ú…V¼¦a…’ö"Óep¥Æ‹¡4—ˆúó¢%ÁÊ›”bV(#ÝB5íÉ‚»ýƒ‡„ÜL“§6‡Ø”CÕIê$|À­m¢NB,¯±± ԹĪ©{¡¡*3˜ò*~DÁby~.Ï‚2nD(7O‚JºpXveibôY{ñª*´–Béƒe™ûfB©ÂàJàH$*ypœ,r¶r –·3°niFÄXI*b³±_Ì âù€øÁi$‡HëòU®êUDôˆ!ŽàCØ$Ó„Ÿõ>¦™æ.*=i˜@'ÌA\Š - Étl"¬šö4ö«´šp¬u)q!¤@À¬ƒ^Î…Ç™P‡i]*|œ Q¥sÞIUn„™iá§Ðã›rPÊ!ºœvjå®È%xb<:ÿ° (é˜ÿ§eò2aH8ª²9 ñ?-KÉÒdŒIÿÈkßébvFl9cZ–ÓZgÇÎà8MU‡o€ã)˸€îOØ9HØ_µ™JE ïëÞ^q_ÂÎ'E&ä¼×c‹›'/tõx‚㨣x=Øžvþ_aÛ¾‚3.ä©ë}…$hrÈ'_²œ(âbAZlìŸ&ë6ObíYb£#g.”#²GZ}\S‰Ìˆ²EÛyjd„'7 ŠYh9/ÉHÄŠˆÉÕr1Ô vPˆC’Í%¦l…”vZ ¼ŠH¼Ç…HdQµ£(£êTÚŽt%!!{þeGV8eG§¢ÓRR¥¾õM1b+8˜sªâlé4êÉANÙ:⨅¸ÂšSÄùÖAá8–+m,p‚êM Fs¥žÀ°àçÚ,«"¥›ªŸƒH!MfZMÓÏ…ÑÏQxžŒ£Ó†S#¤J ?3¸/wž²¡(ô9”ï­—ãñK–üxPõ½N‰É€3g]î{³ôÛ•â̧„·INú”ª|Èãl’~ˆÀ9v)¿¸0³Z¡õˆOfþÀe²§0Pj,¯u Uã”ÔUãS5¥\?\ ¾ ÀÊôk ø’ ð§û4ŠË!)¸ýC¸©oá—²É*u$ËF0Ŷhš%á&Ÿ±kŸ·ýR®»)U”¬IèæÇZa,Ýv€Ï"IŒ¶z®e$’Ä5L™–=Ɇ±~·¦CüÀeO·x€=•=-ਵަì D#ÖÖÝeOê+{Ú<¶O‡9ßLb'_ÚÞl+‰fâ×.xªfÁôHr­A«ÍºŠçq.yªl^ñ\ë%}åxýÈ þ!2ìæL ¾o¬üPøÍÊP£7vhœæ*CcM½s¯BÇêjçc€:'$veI)N©~ÒǧÉF¥”Ü ¡¨+„¢ÇÕ6Ÿårzî•ÎÐç\Û=H9–"Ö»å\N«kœ›BÕõÈüAΛiåO¼¾IËÏ9Õ7ãŠú&çA,–4«ê›\pÁŽ¬ÔðvÔ_ß4—¦¼gõMw¯úæܱºY*NbZä I¯P©é4:AÒu(’»CÏú9MÆUÓ‚&S›GÇ$]Ó1é=CÊ;ƒ[é¹®HÏ“rW™|aÊäâüWýÞÙËzW1 å°Vƒ$Ùzµº. gÊZ)×%cµ.¹+·f’ñÜÍm• mñ¼SÙÒo\¶|PzD]”¹¹ÀÚ:¯ˆ(äP^Gl…C—Çpi\³´M|rÍÒ¶îÐe5ËÝv5Ë‚/7’@RÔƒ½Žš%ÈÜ~ÍŠ–Ü¥¸h‰H#ÄõEË44[ê‡5]æA18˜”V¯ë‡åÂ¥l 9@¶§pÙS· ðIÝrÞ‚“™Š?ø¡8Õ.°•D~{[ññ¬í_g†0š‹W§~ñ.I^å÷¾j1Ëw b¢s«ñ¾j5 ­é‡6¢ý[#¬©Ç^ mUªÿ†áPåú+ì€ wÀž>„ ¿¿¹yõfO—¯®n÷~˜îéÞüôóÛì”®·uß¿+ªC#ËC¦Èóo=ú-]Þ-¯ Ñ–0ä×ÝÒ%NS¹Åq’ézm±yIW—.é6‚ñÚ-²Æ=ÄéYíä iœæ>½SfüCÑ%¾Ô‘·¯7#´¸þžeÍIÞw{»RcLÀºmÂ<[7öÀyìˆ/ezn Ò%·[W·QNkQ¾}ƒ±‚‡ºn0ªa~Uê‚ÛGkn0ÖG;®[¨<4o0ÂAïT˜bÛ+Œªæ*´æ&È ±–Ÿ4öï»›Û‹2ÆóÙ?jì*¤V•yeø²%¯s;wËŸq ÜÒ}qк…æ«®^Êñʈ–ÈÃ[y Mj6Š¼øÿKh]B£ñÂ…íÒD3è梟RªÐv«KRÁ¸‚§Ïô:ÞëÄ¿þtYÉÈR Ì/C`’‘åH9Ž¬™Æ–‚åÄk N*Ý#g…¤Ÿê•V¨"PùY¯‚ÎÞÛ€_ÆüV§+d( îBÔ8Uü\bfDx죰c¨~ˆ&]–ϦÜxœ‹µoÂòɆå£×%.;"¨&.†%ã’¸L– ̵y[¾ÕË^xöI‘@hçÞÏèƒítvvW¥8½äcL'}o:é»ÒɺÅE©ïb¶v:i¸þ(e¶Ëê01½3Ë/8à·‘h°¹¢û2[ÆWÜi¾T´q÷IïÏO9’äHÍfe)î}ˆ7͹xv~žqþ‰²²~E]°‰}Áú5o#°RÜ:úšæòSËôe42¥Ø%º6}3Þv]]‘¾\ƒ¾Üvôe.£¯©†e‼Ë6È 0=Ž/Yü&Þ£’”¢V“*â¼ó¬r}±Îöëb90!©zDLÜSQ/rã©lU÷—pF®Æ B´Ó0W]ë"â—Ø$*/Y³4"àwÂtäê-ð|îTEÓ4!ÀÑ÷êB6©5ر*=·ö³lÐŽ_©Žª@G `j3rƒœ*«EÞEò¥”¡C\É»‡¹$‡ !¡Ä»\Ò=šB^Ù¸S<ÍÅš 'EUz‚K‘ãYXe!‹(5ª“ñoPZ×>Œ­ =æXÙÔ«+qdûûC8EÜ7d6ÍÈ' &×)’Ì­žña^µÊ/ľÕCˆ•ß_§5b¾ÃoÿxáÍóCiÿêíÝ»ý§Ûëë×eí9ðÀ¾[ÄÅ®œ©È¯‚‰;ø&N—¶Ðßq®ÇÔ_~: "]Ë€ÇÇõê‰ ‚B¢”f¶;Y}dÓùÖIZýÖIÆòöù™MÖí~V=¦òùœtî\Û—+á.ü;_SèÑ´GlYož ûj+ZZ÷VÆ»¡æ*²JÚ¼Ti¢g%cbv”äv¾d3ûç ýb§¤”Ïÿþ"Òó«_o®Þß¼»½:¼´ûË«ÿ\ß•ž‰åÃöÍ”vσŠ¥PQh ãµÕ7åŽGìE xßIez'õìòÕÔ€˜;~òG]Mï"¿y—¿”ü.}…yòRó…Nˆd#Ê”K”¬qÝn©µÛd¦?Ö2½Ð`Õ°ì$L/Ú¿ù8†õ?Fizƒ endstream endobj 840 0 obj << /Type /Page /Contents 841 0 R /Resources 839 0 R /MediaBox [0 0 612 792] /Parent 838 0 R >> endobj 842 0 obj << /D [840 0 R /XYZ 125.795 791.866 null] >> endobj 86 0 obj << /D [840 0 R /XYZ 72 137.809 null] >> endobj 839 0 obj << /Font << /F57 146 0 R /F70 201 0 R /F79 203 0 R /F74 202 0 R /F81 206 0 R /F83 207 0 R /F77 565 0 R /F80 204 0 R /F73 223 0 R /F59 148 0 R >> /ProcSet [ /PDF /Text ] >> endobj 845 0 obj << /Length 3495 /Filter /FlateDecode >> stream xÚå\I“·¾Ï¯è[¨’ˆ`_’JÊN*N•Ë—”uòr`† 9æp$ëßçzaDo$G²œ‹¦IB¯·~ï=´xWÐâŸ7´úû··7üÆЂQâ¨cÅÛ»‚sC”a…aŠ8S¼]?Îî_ýüö[ tíš«@%Œø‰*Z Š¨IÂ¥<b*OI Ñ¢Ä^Í“³¿”c•i†- ükuù¸vó·7¿Ü0Áë n,±Ra –#‹Û‡›¦Å?~[P"….>„¡X-ž6Å÷7ÿÊò£¦E-‘RÔ“¢™Ÿh½B¦Dkf '[Ch1W W1v›ã™%L+ÿ笟qŽÐ0ù0èuŽ]xÔ†ÕC6Û*Ξ 2& ’禭 Õ-}èd$Šê^Âcö$ÚµäD;ÙVÎEÍxZ ªÒLücvNÒæ&•è±ã®˜·†ebE¼BèrkÜ<ûJgy='–{ß*@“Å\bí°MËûªrÍ1 Z eš·†õY¬x!‹Uƒ«­ƒ•™«X¬6,4£,–OµX—X™²®4Y‰=ÑEo|”2ªqʨF*£©Œ<¯Œø`¬)æœÕvŠ2fµQû<Úȉdz¼#»DHâª2j¼¸ÈÊÈ2¢½¿àŒÐc˜b2Ç08؈a/Åú˜`Ìúü€y36Mi+ª03FœòÆ †pîÕÌ{PGŒ«8"‰y5g”ÒÙ÷‡çåj{øÃ+ÇgOˆÆÎåŸïÖÿ]mÖ÷»Ý2§@p  ªÐüö~õj.¤˜mŽÿ Ÿåì~ñTþpø°+î?>®ö‹ýâauXí«_=÷ÿ½;Ü—ÿi¿zܯž0­Õ²üy½­ˆ”¯‘³Íî]ùÍr÷°h~Ý•?®¶OÏûj>»§õaý~}øèÅò'¿?wï=+¾š\ËÙrõn¿Zy&h1ÛÝ•_Þá+¼#ïÆ<ðÐG±s“ã<9k”v±]–t«·ŠÙÓíb³ÊÚ4ÀdD䕽ãžCå…Žür:+uÑõaí9í§ñ°ZdÁ—òJšqÑUÑ^¼¢w»}ŽGÜR<âôËÌþš#ŠÁV5º5˜hÅ81{¿Ø¯ÛÛU†–N½KâÎÅ2‰£¬”ŽhLIA¸²½Q>©$̶ßÙÐÒ…c©*D6øyt_çÅ E47ÕF‹Ëõ"@ÈÖ Ÿ¸æ%·ûÄ Žh}–8¹'ɺm¡Dp&s¡kªù<æ]­—Ï”|«Ò^›D7Ù€ýÿd×IoHÜåfßlyìô½ßÌ:}d`J5²ÿ¡"b{2Gb´ræЗHúÒó×ÿ± ÿyŽñ–PÚCYn Hæ\”9ùØaµiãÔQL¾ËXJu¢ð(ÈIÜF\JË‚ŸBk͉O4Lƒ­=FïuЊ~l $¨…jY}¾öÑ\¨³ v¬ Ü ²@L×G _îB~(cîËÂ×3¼…… ÛxgÂdÆm‡Év;¦×Ç 8ÊD($f½À¬^¢¤ P$a½`îèo9¨²˜õÑ»€à¼ÄÉ8WMr~x6åm€ÈVÌz“ÕA$šfšØÚ¬0kàPÁ€íÇ„¼Fý± 1Ôa c"˜¸p渇Κ©ñaåõ(”˜ÈÇ£xm–À…`ÅÄx½KRÈ FfŠ]$D‘9°“ó´Ñ±×K&gQP¯v«½·ë3Û†—!Ž^MN,+§d!Ó /7øÛ»/QÞÃdañR åã«=lö\61—ñ“…MÂIÖ · Rò¹¼·6`×7c¨¿ nº“÷ ©˜÷Áe².·‰˜KWK´1Òq ¾€e2¤î`Y“VåAß-7‹ÛU;úúñq¿ûuý°8¬wÛ¾¨réß0͹ä¶ÇÃÃnY=ëüpXoßùÈJö»ÛÕIÊÿÙÍ>Tö+Ÿñø§åj¿~7¿ ¾x~¬ˆïê—¬÷Ë’ÚnÑƳ’ÇwÇiÖ)ÑV"$+C]T™_IÕüñ´_ýò¼ÞWžvõ€õ¯aÂxlÑ&>õR³¯7›ò'ŸŸù¿‹ÍæMùôüäY‘s£^³]IöyTaŽym1‚·ô¢ Ø5• É]v´`µ´kîïCŽQ.Ô»æaD˜¤¾Ók_þYo³ø~·œÇÀ1=G8Wņox@O”AÙáB_ åO^Jr¿Ú[¹õàÍé÷Þæ½Ô^"'ªÏ½#S„O¼ÓI–âÁ #A+J æz%§¢2#EÅ ³MˆçY}õ#­¡¶7‚®åcx 4&T‹=-YŒ` ÀƒH¤õ±¨Ð‘K9œ·×;ï¦u†¥»1™pÄbA£ÖZj»×ZÓJ×ê{^X–hÅË´/ío9"€¤[ ®y‡§Oxù‚6.)à–/ì_ÃÆ=Ò6'f‰u2'•ÄøK©¼h9 ÓFT.ê;ïóÑ«‡ù| G@¹c˜I Áº\ÖÄÀLÙÖ„¾|\žÕ¤Žeà›Ô:HÛ¸ÿ›”¸òLî!(Þ@+oýUŽ{ø›)±2ƒÜ™BÚ®r;Äkt•÷Bw¦ p¸åÄ43Ïbî“¥îÊ ˆN¤éû°D‡×M5ÞSäuæU¶ãŒì„"µÊ_ûL°G¢,-Ã\ ¥Xo-©&îêÚ:;­J¤VuºJ¡y¿ G¬g%ü¢­ˆ‰>™2Ýp:Õý(™¦>ás}•2לGk#Òˆ(¼~ÛÞüÐÝ×3b1ÆøJñP¹ŠÃÏó–ä¥äX¶ðÙª¶„ædF@1)¥ðYõEe(åñ£B/»·[¥RrÄŽSË°'¯¿½K«¥¼²^Âp¬‘½ìlf«n £þ¬§”CÍìšÖ‰RžÆG^îõ]S£ùü5»˜`UPžË—äRÅL¤TãÀ¹ÂL)Ú½hÄ®}KS[èšx`ÍÙ…Ä¡-à§5N¸= $Ú…Äz‰ÞA‰«ô ªƒ‹ÇQ«óñëø˜–ÇȲOË~/X=µŸ¤u”Lçʇe²GŒ°É©¾CÙÂMiŠŒ¨£ªî:*ÓÄŠQž‡ To%ÕÍfÁõP'~וT3õ þù•TdWJ³+VRMoÿWÇ© à·¯äwpöŠ©‰1§Glp±ÓÂÌÃM|>"#Þæðï—¹#…Ýxò]æ¿kÅ Ûþnv­œ\g½Í`LŸ/½Æ>’"k"µÌö"Q6ˆš‘@´×øýÁ'L†!Ar¶÷Ø–áôD­‰i_„´¢Áç÷'¡ý·WG4¾ûn¥™[λœfÚ½/Üj×C®ÖÍ\Zc¹$“gìs¾°)fˆ0xÿ—0¢Õ^ó.áÿ{#’v“zí<öþÉ…Òw4\ØGÌj„(‰­²ýýùáyÎaU×xìÞ­ŸëÛs/õàŽG—zøÏáRÿ°ÛV#NN›9DÝäá¿ gÁ\}FÎÓö7yø‡ò&Ï’Ê©ÙÄ6“áO‡Åv¹ØWÔ–«÷ëÅ/“^œÈ¶·Ä9ûÞëM@úû35-mˆ¶f ‹Å©k÷_Ô)Ðc®úí‹ý…dâä*)µ×'ªÒ$×,ôNìXÑv_t ‡«‡+© G˯!ØšV{áò¥k†äŠ5é¯ô»\°5©É6W”ú@Ä ¹64?KõfÑ(ÌïrêÍï endstream endobj 844 0 obj << /Type /Page /Contents 845 0 R /Resources 843 0 R /MediaBox [0 0 612 792] /Parent 838 0 R >> endobj 846 0 obj << /D [844 0 R /XYZ 71 791.866 null] >> endobj 90 0 obj << /D [844 0 R /XYZ 72 627.345 null] >> endobj 94 0 obj << /D [844 0 R /XYZ 72 505.192 null] >> endobj 98 0 obj << /D [844 0 R /XYZ 72 147.257 null] >> endobj 843 0 obj << /Font << /F70 201 0 R /F79 203 0 R /F57 146 0 R /F74 202 0 R /F83 207 0 R /F77 565 0 R /F59 148 0 R /F80 204 0 R /F73 223 0 R /F81 206 0 R >> /ProcSet [ /PDF /Text ] >> endobj 849 0 obj << /Length 1098 /Filter /FlateDecode >> stream xÚíYKE¾ûWô ¯Ð4õèç )â‚ð-Éaµöfyíeƒ²áßSóðxzÜ3[ÎB 'írUuõ÷}UÝõNúyÍë‹Ù7?Ù¨t„ˆjq«<)oXƒ±j±T¯çFWÀü—ëÇõõÍê««Hó÷Wù0ÿþññiûñþáúÏûíæêíâ•øó]  rÚX¬Ý-îVõo·ëõöù~ó®~»\=Ý'Vãë§Æp³Z-WËúùùn%WQ“ËÝip¶“;7F¦ä¤‡]ê‚íd+³MŽVNÌKÚ˜ˆ¤Ì~\Ìþ˜¡¨Rq§1*¤QÔíæaöú-¨¥|÷JA…¡çÊò¡Mk­~›ýÚÈaoó*W.’†ÐäÅùº' €›"±d“'¶ oó2 о »óØcéÙC_cw±“5c.@A(*åXettT[—Û  Jð1mJ‡±„+çI;Â1È‘ÕRÈõ‚Ũ¬Ý9Ñ!ÁIy¶ÐE]YV`Bc—U ¢û'áÍt0ç;•b¶¥üA3@^£èJ!Æä¦4Û6éÌ÷AyD9'LeàÿnHCke¬¶5)^€ƒa"Ý­œØ¤gY9¼:Énìü ùCåàΙGÍÖœpõ08¬† ê;ó¬ÏZ{£æ¨ŸM=üËõpá}ðÒé½;ÃvaºØ6Þ`à°¥¬E]â3Ž€y)rÇæ€Z„ü*¡ïB²È¯þ\ëpìi|ºt“ÁV[íTj=—zr+ü0p]á|5ïŒÕåã;@vÊ”¾EÊZÔ!Œ^U——Å|yØpÒß /{51šÊäÖ2tÑNœoÃ…ýa°±„$lsIØê—íÿqªp,—cˆ•RyïP#“+¡òß!e+0 endstream endobj 848 0 obj << /Type /Page /Contents 849 0 R /Resources 847 0 R /MediaBox [0 0 612 792] /Parent 838 0 R >> endobj 850 0 obj << /D [848 0 R /XYZ 125.795 791.866 null] >> endobj 102 0 obj << /D [848 0 R /XYZ 72 754.004 null] >> endobj 847 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F70 201 0 R /F79 203 0 R /F73 223 0 R /F80 204 0 R /F74 202 0 R /F77 565 0 R /F81 206 0 R >> /ProcSet [ /PDF /Text ] >> endobj 855 0 obj << /Length 3333 /Filter /FlateDecode >> stream xÚí]msܶþ®_Áz¦º1ñž™Ì4mOÒ$uÍ4S;è;*ºñ½(<Êrúë .@äQ}Ú;ñR¨;p ,ö Ï.x$ú5"Ñó3âÿþíâ쳯…‰(O—itq©4Rœ%„‹èb½ŒÅÓsJ‰¿Ï³õÓóTéøë›õ¬\lÖÛ§¿\|koW%‰!†V·“蜑„íîþÒÞbd¼r7_ú›Ý½Š„÷ê4á”YpëÊu¡ŒE*1Š@“âsšhžºn¯“ž˜ ‰Qš-£ çç®›îÎ+ µ“ïÊå:ÔðäÓÄ´RþÅÒKS‰äù´•su:Ž°  ³ë¿ðœJÒ$Ÿ`-µPÔ’°a[ªü°ÙÐÔ|O?¢+×EÒ;µ¹2ÝMœž´bÐ>\ËDêÆ;?ÕÚœö¡‹W$ý÷ú@¢7Q}õ75ÛßêLÍØkˆìÀ: ´Ç”DD Šê_wù#°Htì`Thüy÷½<—ÜîGžÚõ‹sh3h×(¤¥´¡Ç9×ñm•%Zµ3ø¤„ë´›æ[oQ,L ŠkI£°Hǯ\Ž3º´b3ñZ4¢nŒ¿cqð9,Kmµ€²¯½ÞK—Áç8Ò¢H| ä h7ÐΠ͛ÇK$ QDS¢©áa*Mˆ¦õ`KZñs·§Ô©øÂ>˜[ûVµð†#‰Œ¤Ú J°äP‚ê™øˆvpŒ©ô6®c*Ói˜ÊÊëÂïl›£Ú>É¡)qVÊÄo`ù°ükhçÐn±V "·U|89~8Ÿ¢ÅÕOqŠ«w|ʇ(îç8Š – ŠN0m"^B‹¤¾©Wß5´s¢Ü;hÇÀ´×Á'ÛÀ:–^ôDZ—õØ»¯1`¸Æ`ßað‰„oG¶GÓ臹b‘,r~C,_äîå9ÕDx³á¢¶,X·ÿŸÞN|DLا£ˆ*Sñ-( Ý`cÞ†©m˜ÜÙ†õcù}F†¤:£&I•Àÿ3¸Î?NS†%£B9 ð¢zshé”;ÒéÌ¿ìtÒ©OC:e`GU`Guü÷Üg·qï¸Rc7#8’jEhÖlõ[×ÞÚ¥ #%ceåQäØœ’»}±îÈñwÍŽ­ŽûÚ>:.Ž Í̤XÒL|Àíñµ¾«6ƒ¶À3q²,–ö$Ø%>P„ý#ëp>²¦w@˜f“Þ\ŸlŒcüö{Ê‚ëømG ÛMXÖ¨ªö8ÒãÓƒÀ±£^çƒÇhHÙã…VRx£÷®Yl –$ä¶ñ×e`07þ[D(X ¹Y‘zø 1Ã!!ÃÁL\+êvã±Q«¨‰¯²ÓýÇŠ¥lÅêq°Ç·b l |NÖh‚:Æ 3°%“›îHX­³Ó2Tê$Ù"H¡­±v€˜VY·ÓH_‡NIî8%‰ñÓÐ໓äÝžpâœ/À'¶½m²®)tGN²}T¡a§¡Bw£ä/š\rÍ™ùI¡äœú-Kl\æÆî)·˜ÞIžXþú®Žj‚ûÝÎÞz!OݵüÔˆëOȵ¸;Ô}ÙÖÉÖKTQ+7Äñ«ïªÜЕüÜQ§î¾ŠíÒÿæÔò¦«@žœ.¼ÁB.¶> ÛªVŽµ¿šàSb¯{`œ~Á~ÀÌP5›‚šTLF`Q½°¥¿FñWz" Š™4ÌMüC'¬vr ¿bUáÎzXk çøÛGeÇ‘¨UvÌ—¿-Aª—'’çÞßý= t ¿ÌW_…šA»‚êÔÜ×¥"ÕÜ?‰Êë=×ÌÝûÙÈžfÿ§(ÿ,y2òYt,–m{> ¼¯­KÉ4lnJ'2Žt¶ß*¯ËWÔ‡1¶ ÿ^Žû k ±Æ<|£|T î—¬±ZÔ—Ð✙ß|€šÞÚ½£×ôÊÎy~ÞsJ¬: € _(L­}ÉÏËÈv…)›ˆ9È?e °`Ä9~¸&“4b]´ò´_Â'nïüì;ø”ᙪO£rÒ+ÒE0~€O¾ ÐŒg£õý 4–#Ö Wïõ貎¿º·6‹‰X9‘]+«1ºJªFÞ§î;î"@¼Ëå.ô{‹Z Ç(¥/n< A"T\^í"à2Ψ%ó(Û¤þWêkyÇ‹\m fÇ+Ànr‘ +rÐ@ÿIV⮽¹¯{–õµ ¾ÏK%3f^cU¨- Þö6·j"æÖU,P(&tiXê?™A[B;òü%:þkwã+Wí¢Ùq,£mÅ!Ȇ­Á},,tTª¸Þ—‡;RÉÇHtÍ*ÊW#{²Q4G¡N7·ÓM|Hwþ¾FmûÛYÝyK¦P ×á¹J]læ;€Ž,à/ó¤=ýjŒj[A° S=™@ûë ä@ËTÇdЮײôæØîÄø“cÙ‚9Î<ÛZíœbêŸtª ćžtrôŸ`ñïx[Éáši°ýÛ~XødÜÚX‹üÕhÓó FΦÆ1Í4F&“+IO'WbÞó¶$t÷Ëð^GÔÇõõ®¯›ƒ»cÇ4N$ƒÈÒSA÷WptÿC±d˜£€%§c±dmBËÈw£÷]hð5cÓ8Ëø¾¶Ñý‡"êé-ž+–Ák½êA´Ø¨Â û2`êëÉØ#®˜“Ë]§êC†³²{{è÷Rð•‡|ÙåyØÛ‹‰¼‹’Ü´IAÚ¤@“&ÈÀß%VA" ì¦0“øÖÅÊñ»)"˜.þaÉNÀ»¡a¾s˜éQ[èQðÆ*(•ÞÍP⬲иP,T^3·Ö 3ˆ~™Ÿ¯¡X›ÏÏQm¼é¯&™,8È›+îJ„IÞ—,ØÛ•ÈÉ`2“?Êb£Ž5a‡KZ8@†¡$-¶­Áf óxPlen‚w}®ƒÚ°ù¡^çÆ°‚i~”`ZhìHù=ÞÛùdÍ»O‘Š¥'„XRÄLM‡Ô“ñìdÐy4;Ý"ˆ…Ѓ'Êà | H½šRo¦9rrH½‚ß ` ¯×Xx= ÞÛ ¡ö[(ˆfÒGÂëS:¯—/E°8~Š€O$³ÆÓ)‚)‚?bŠmƒÏ™˜b¡Ký½•<ÄÞ y3¤ýï“e0•AÑY]Œ†ôBÜyP&Voy{Çû`ÃÂdœWFqö>úÿÉëæ;6œHF÷,F×™§Y¢Òe‘÷>WHFõÁÜBÈYŸüÈë³›ªì%Ï/…©ÊðüôA2â®õ&xeÏë@• ¬t×¾‹¸·g@¯Ï ~8üYvªà×æ«Ÿ‰§RžÜvqö?¾FB endstream endobj 854 0 obj << /Type /Page /Contents 855 0 R /Resources 853 0 R /MediaBox [0 0 612 792] /Parent 838 0 R /Annots [ 851 0 R 852 0 R ] >> endobj 851 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 589.738 51.396 598.177] /A << /S /GoTo /D (noweb.NW2i2LRH-1KNHNk-1) >> >> endobj 852 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [161.931 587.1 173.894 599.996] /A << /S /GoTo /D (noweb.NW2i2LRH-1KNHNk-1) >> >> endobj 856 0 obj << /D [854 0 R /XYZ 71 791.866 null] >> endobj 106 0 obj << /D [854 0 R /XYZ 72 754.004 null] >> endobj 110 0 obj << /D [854 0 R /XYZ 72 680.789 null] >> endobj 359 0 obj << /D [854 0 R /XYZ 81.963 590.823 null] >> endobj 857 0 obj << /D [854 0 R /XYZ 72 588.096 null] >> endobj 858 0 obj << /D [854 0 R /XYZ 83.504 590.823 null] >> endobj 859 0 obj << /D [854 0 R /XYZ 83.504 578.868 null] >> endobj 860 0 obj << /D [854 0 R /XYZ 83.504 566.913 null] >> endobj 861 0 obj << /D [854 0 R /XYZ 83.504 554.958 null] >> endobj 862 0 obj << /D [854 0 R /XYZ 83.504 543.003 null] >> endobj 863 0 obj << /D [854 0 R /XYZ 83.504 531.048 null] >> endobj 864 0 obj << /D [854 0 R /XYZ 83.504 519.092 null] >> endobj 865 0 obj << /D [854 0 R /XYZ 83.504 507.137 null] >> endobj 866 0 obj << /D [854 0 R /XYZ 83.504 495.182 null] >> endobj 867 0 obj << /D [854 0 R /XYZ 83.504 483.227 null] >> endobj 868 0 obj << /D [854 0 R /XYZ 83.504 471.272 null] >> endobj 869 0 obj << /D [854 0 R /XYZ 83.504 459.317 null] >> endobj 870 0 obj << /D [854 0 R /XYZ 83.504 447.361 null] >> endobj 871 0 obj << /D [854 0 R /XYZ 83.504 435.406 null] >> endobj 872 0 obj << /D [854 0 R /XYZ 83.504 423.451 null] >> endobj 873 0 obj << /D [854 0 R /XYZ 83.504 411.496 null] >> endobj 874 0 obj << /D [854 0 R /XYZ 83.504 399.541 null] >> endobj 875 0 obj << /D [854 0 R /XYZ 83.504 387.586 null] >> endobj 876 0 obj << /D [854 0 R /XYZ 83.504 375.63 null] >> endobj 877 0 obj << /D [854 0 R /XYZ 83.504 363.675 null] >> endobj 878 0 obj << /D [854 0 R /XYZ 83.504 351.72 null] >> endobj 879 0 obj << /D [854 0 R /XYZ 83.504 339.765 null] >> endobj 880 0 obj << /D [854 0 R /XYZ 83.504 327.81 null] >> endobj 881 0 obj << /D [854 0 R /XYZ 83.504 315.855 null] >> endobj 882 0 obj << /D [854 0 R /XYZ 83.504 303.899 null] >> endobj 883 0 obj << /D [854 0 R /XYZ 83.504 291.944 null] >> endobj 884 0 obj << /D [854 0 R /XYZ 83.504 279.989 null] >> endobj 885 0 obj << /D [854 0 R /XYZ 83.504 268.034 null] >> endobj 886 0 obj << /D [854 0 R /XYZ 83.504 256.079 null] >> endobj 887 0 obj << /D [854 0 R /XYZ 83.504 244.124 null] >> endobj 888 0 obj << /D [854 0 R /XYZ 83.504 232.168 null] >> endobj 889 0 obj << /D [854 0 R /XYZ 83.504 220.213 null] >> endobj 890 0 obj << /D [854 0 R /XYZ 83.504 208.258 null] >> endobj 891 0 obj << /D [854 0 R /XYZ 83.504 196.303 null] >> endobj 892 0 obj << /D [854 0 R /XYZ 83.504 184.348 null] >> endobj 893 0 obj << /D [854 0 R /XYZ 83.504 172.393 null] >> endobj 894 0 obj << /D [854 0 R /XYZ 83.504 160.437 null] >> endobj 895 0 obj << /D [854 0 R /XYZ 83.504 148.482 null] >> endobj 896 0 obj << /D [854 0 R /XYZ 83.504 136.527 null] >> endobj 897 0 obj << /D [854 0 R /XYZ 83.504 124.572 null] >> endobj 898 0 obj << /D [854 0 R /XYZ 83.504 112.617 null] >> endobj 899 0 obj << /D [854 0 R /XYZ 83.504 100.662 null] >> endobj 900 0 obj << /D [854 0 R /XYZ 83.504 88.706 null] >> endobj 901 0 obj << /D [854 0 R /XYZ 83.504 76.751 null] >> endobj 853 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F70 201 0 R /F133 581 0 R /F79 203 0 R /F80 204 0 R /F73 223 0 R /F33 580 0 R /F136 582 0 R /F18 190 0 R /F74 202 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 908 0 obj << /Length 3706 /Filter /FlateDecode >> stream xÚÕ\m“Û¶þî_¡~è„×øpÄ;ÑÔ™¤qœIš4NìN;qü–(IT()ç›Ôÿ½‹’ ¤$ŠÛs3 ‚» ày‹%qñäõ$ž|õ v¿Žúï—Iuöþàê —…” bò|>1Æ&’Q3>y>›¼ˆ¿xùü›ZPð«‹ tØf® 6 r2]>xñ2žÌàÞ7“QˆnLÍå„aŽ„ÐëbòìÁ F,Q}2 ¶mŒQ­TZ~†XËâà>©mô3ç?u;–uœÒÏ—اÞYÐîc= è `“~‡Ç –ƒœáSÇ©=mVã$1x9|7eÞi?f.‚‡*ÞÇÃ-Í:zuÁT[5Óâb>TõÞ ‰¡ÚVý»büGvêUU#µS2ƒçiµóçÇ=Ð>Yo–~Ï‚¶æ çG6‡y„»•ÓOYYô{˜Ú`ûzw÷ |s­ƒ`}•.nÒÛMÞÇ°¨Æ»oQܾaJõŠ#>_é'=È;Yi5°J8y&ÆÇA£Hºö÷«¬wÔvàÎÓÚÀ.ûu—.èpÞÆ°¤Æ›g îÛ0pótö¡ídA¤0&ÎC ðÀ(h# IáB­/ žÚön˜ÞÀùŠÃˆK«ôX•®¶=øfXÇüî6íà(J+ÞQºÕ¦Ç:^p+JÛ#Šh“Ÿü±Æ'[„ SgF¢°à“±obd úVÈQràÝPBI51Öo/úÓ{”Â&5‰^_XÂÈÝô}+U§ï“DyqѾ:“‰ò*¹òìÈÞ×bsÊàhðÉm#ï_Û.±BÁ1†ÓÆO¾ ñYk¬‡Xþ)˜°¦¾£ôn§_8LA>rqž_` ;V£»àWBì¸|Õ@Åá(¡¥ù`¾j êp<ÛRý~å«Æi§›X[íìÏW®7Kq^ºŠ =K“q-1xHé2æÓtÑ“7¨·ؾâQì¯ÝÌLÙóa–tàÜ·ÆŸ^`ÿ}¸Ú÷·äñ¦5ÔGijõ;M Çù¤yÖGÊ#VŒ§[D»ô<ÆpŠ=”íƒ>b‡ Ã!©¢‰g»e])í`‹¯5¹·w»uoXÙÁAf%$H!ëOOË ßÛO¥hœ´¿ i"8Ñ Ùá{”¦êØp^±5Þ=‰Å“u©AèyYlFˆø(ôbÍ0OËbÖC¯AJ;èåi]îÛ|]%`N£Ö “:¨¦%g~8@-y µ˜ú€¹5¨g;¸å vµNU¤…ì—p˜Z„ƒ9N¬GÂØe&ž7=ܦµƒ\¾Ú2Í7HÏ~lR´n¬‹›¬ì!Û0;b@ßÆ®àÏ2$¾§DB‹Œû_Íz¨1°ÂÜðû¡'+ ›¡„¹Æן~©‘Öøºi¢Zi§›_zè1Lm=|½ét[çúÝ;¬åºXe«mð{%En>Zþ:ÍKD‰º›´—ª;iÿ"”ÙÒùÆ×Ø¥¶ˆùBÁ<%À˜Zùãp Áû2¤¸5Kóð+$)>øD~"Æz’œœd¿ƒŒ˜”þ›†bQI·?Êèƒr|ØŽ;´;ô㎯É\ήwô {#ìýÞèóŠ'+ x4PŠO¬ß5›8`½­·3°˜Uû7¨Û¿ñ¯MúÚ;À$ó;Ò–ŽgÅR Ê}¯Ï²7©¾±WÛô—leOÍ~†ÀGõ˜3$hí0fÅôj§-Ö&…¿§O$bM<_,v›m™n×ãXÝ…ýÝmÜêÛz}n¶šXË›—5BÕ)(h(? $÷¿Î÷¶Ÿø±‹ùô~~Û d¾zê"è™…y {˜F°‹ïŸÝ®¶àúóiºŠ+¢_C¤Jk°÷ ®MX³…eô*ÌÏÜä›ê5³ýòßAàÕmHþ¥~-¥—$-Ó—`ÌÑ_é_}¤MüÈ–ýÏþ|¦‹ìÝ«+[äúßï U›Ý’÷ûrõ¶% º[ë´L—mñrþö­-Þ«§Ïæî =OµjKg‘ð?Ðk—DE!ÚUç(ÓâHB¢|ë~u&8ÊòíµÞÕ¤ËR[¸´_UCÑ´€áÎW¯á†Þ›b‹ =›§  BÀCï>X9—N\qàY§ÂX¬šðšg, ÒG…Þ¬²4* Ð|7NM¯½ÊWn—‰¾q-··R[0Íôí¦ ,Ó[[heW{Lt‰O']×pÊS§(Ír}?]ØòE¾ÙÚ;• tÂumuó÷Ï]R˜¦%6àãÄ-  Mû®æý{]‹Ý.¿Õ(·•$ÍG9,±¨Fis°íŽÚ¡Ch‰Y{‹"°S2F2yã.'f#lⶢ.ͱ0G½ÝH™ ¬àOͱ¬·«V%ãlô•±Ù-K2f0³[–µ“X˜¾<š#Ž™Ž"¦Ó¨Û¤¼E4Dïs³xWÖmçæ¼p[†uÉæô-³H-<’qñx².¨¦ï->´ýìÂRWy;Ê•#ðsÜÕ%‰“2pRNÊÀIáÔ½¿/ûÐ`fïÌÎ Þ» =FÍâõîB?¦yJ7Kµ›wg³?ž æzÙœ{YS½Þ`¨]üÂ9úÊ•²¨ÔÓ)åãíÐsÓ@nÈM3¹ñ|$¯Ï”ý.Fäˆà^@Mî# NÜÿãij»„‘ãÔÿAã>\ërá±ØÛÚ¬‡,¶nú‘õÌ¿›Çm[lËês×4}ÔRÑW×®XìõÒÇŸ¼òÒõÅ‘R²gáÞVó &0+_›Á¼5ǵ9êv‚2ÝNàÀ ׆gM ”Ó±¦jV^”Û^P¦”în| óè¦ÄLº¼¬ÎáXÇô> endobj 902 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 715.648 165.939 727.753] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 903 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 234.127 51.396 242.476] /A << /S /GoTo /D (noweb.NW2i2LRH-4BH61R-1) >> >> endobj 904 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [163.131 231.399 175.094 244.465] /A << /S /GoTo /D (noweb.NW2i2LRH-4BH61R-1) >> >> endobj 905 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 171.86 165.939 183.965] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 909 0 obj << /D [907 0 R /XYZ 125.795 791.866 null] >> endobj 910 0 obj << /D [907 0 R /XYZ 83.504 755.001 null] >> endobj 911 0 obj << /D [907 0 R /XYZ 83.504 743.045 null] >> endobj 912 0 obj << /D [907 0 R /XYZ 83.504 731.09 null] >> endobj 114 0 obj << /D [907 0 R /XYZ 72 704.854 null] >> endobj 118 0 obj << /D [907 0 R /XYZ 72 395.03 null] >> endobj 360 0 obj << /D [907 0 R /XYZ 81.963 235.123 null] >> endobj 913 0 obj << /D [907 0 R /XYZ 72 232.396 null] >> endobj 914 0 obj << /D [907 0 R /XYZ 83.504 235.123 null] >> endobj 915 0 obj << /D [907 0 R /XYZ 83.504 223.168 null] >> endobj 916 0 obj << /D [907 0 R /XYZ 83.504 211.213 null] >> endobj 917 0 obj << /D [907 0 R /XYZ 83.504 199.257 null] >> endobj 918 0 obj << /D [907 0 R /XYZ 83.504 187.302 null] >> endobj 919 0 obj << /D [907 0 R /XYZ 83.504 175.347 null] >> endobj 920 0 obj << /D [907 0 R /XYZ 83.504 163.392 null] >> endobj 921 0 obj << /D [907 0 R /XYZ 83.504 151.437 null] >> endobj 922 0 obj << /D [907 0 R /XYZ 83.504 139.482 null] >> endobj 923 0 obj << /D [907 0 R /XYZ 83.504 127.526 null] >> endobj 924 0 obj << /D [907 0 R /XYZ 83.504 115.571 null] >> endobj 925 0 obj << /D [907 0 R /XYZ 83.504 103.616 null] >> endobj 926 0 obj << /D [907 0 R /XYZ 83.504 91.661 null] >> endobj 927 0 obj << /D [907 0 R /XYZ 83.504 79.706 null] >> endobj 906 0 obj << /Font << /F57 146 0 R /F55 149 0 R /F73 223 0 R /F61 150 0 R /F59 148 0 R /F70 201 0 R /F79 203 0 R /F74 202 0 R /F18 190 0 R /F77 565 0 R /F33 580 0 R /F80 204 0 R /F82 566 0 R >> /ProcSet [ /PDF /Text ] >> endobj 930 0 obj << /Length 1569 /Filter /FlateDecode >> stream xÚíZËnã6Ýç+´)ªc—o‰( ´´(P´Ù¥³pgbL»~$cóï½<¤dÊvY¡‹,‚n4ÍÇå9÷E±ìSƲg,üå$Ý¿ÏYõôëÙ—gßý¬‹Ìö­&»¼ÉLÑWJe…’}¦tvy]å\Ÿ¼ü¥ˆe}õy6wÿõ|pãèhnúRꀾi1Àá·ÙUÏ(™/Î{Òæ£óžbù2É Zä+Tç³4ÃÙ|H«+ó)É"Ÿ@Î §h_ày ¹DË(ÉÄÆ`2åˆ|yŸfè"¿ÁzWXï=ž‡ÑÆõþŠðvјxô¶p× €éqÞ·ºži ³ 1¥kˆM€ŒAks`¹ ?’åÿb¿#¿¿ú9lçß6&:ÔW‡3A"üYë¥D{‘ß…ž-Ç.ÅÖ:¨­àù»óž0´ ·Î  ÊÏ{…οbÇ^¾wí4ï-ͨò5I Ô)hDyн¹‚–dõKÚ[àe 9ƒt“ùGLø>,«~¤¦å'²Åf?$ØçžÓ°H›üoìxd8Ò¯ì‡ë&ŠnLX L0`‚ÁY0Aƒ ÖÂ¥¶èç{ÐÏ€þ¡Ÿ õÏŒ€Yºv-Ñb ëZj†Ü5Bš œs!‹ˆ!ªÁÕ†!¤)bˆ)÷‚æ6ž3´ÍÝ×Ãïï°ò”¢°¤NÕC…”5I×òX¸Î°†ù\6É'#HOû{Èß!çSÈkwаš¤åÚj Xf#"”¨€ÚŽ KºÝFIeMç;(±OÀ ô00OÀ •eÕ‰` g JU§ÊšÇCÈexØnð˜þ¯ÖLb«4òø"kôy`Kø_ûf,æf!º]ÁšƒÊiäÝŸŽܸìe€ÓBÙt®|‰eùåŽ+zÁI¢V½ÑÉÇ%øiR X°çP\”Šë |ùÙ[œ™p–D–ðl)Cøõ®j—Þo؃ç½òô¬vyÉ”‹”ˆ$ãš*:Ðf™&‚ ¼Œu'Iw‰2&ÄAÈ>ŒBvŸè\G©Ú=Z>¥ÚÓgyd^ rµÁ/S ºÄp·!ïNqˆ¹®.ƒu^òú´¹yÙÈ]GÑwówÄ{o)L°üXçýÁàäþ.ÎcMØÿou‹Ï¶ÌߎÍ`Y”ÁV‘×rI–ÐÂêù†÷ö$éLJDpdL¾‡üº#]ŒeÌž ~REŠÎ#^xÿ¸ÛkVŵcìì ijH!Ñü‰ìXVãiËj¤j†ƒø‹iž†heÎ, ©|9†¼\§òó'+|™¨ðUD…/ƒØ¥ˆ _fkÊÖ%/¾ õ:@¡NWN$Ú^¤®·&¬¥;áuôtÊÐHPöéýåV|–Êk^¤Rÿ¾ºF; Ú- v†¹~ƒyëÔ¦¤‡ø®ÝQ™`À{Íh~ª¢'Û*ªà©0mÞ0}ÄùOœ´æo˜~Ó³Pèl‡iöRLW™Kql£ ºMCîTò|ôî‚šÖÕë*å ¥,¹£Õ)b¯“²Š®åñ»}5;O±›üŸªHê쾧REéü•_ £´‚œ@úªÔ(T¢¥3S¨û&ͪY(:øËãyt‘< øÒC•žû>e×Kõ-7%wb=JàUX'÷‚XYSø&*¥>Ô72j«åÍ!”á6dsY‡R!¿Èïrz±u[fÓÊš;äí™ìœ¯†Ù"1³m(ñ<Ôq•³jYîùT¡3À%†ci¹.;çâ/ ªç}šád™IÃ`ùz\àn^¢B'Q•óõ>_”“È2%wÏ¢] Wn×2ËÕÑþ[êWÃr™¼â6Œ®)'QõmUsÄ$åwj_®qßbo ùyË%úÄí£=¾ñ>¶‹°©g¿eÏÖQûQc.[÷짲^K`vš*t»ÅQ®!gõÑ[}Õ2ˆZüó$ê¹lü* $a+‰~îÓÅȨŒýý®y*ÿ__óî1í!`ßq *¨¶F\4âæ/ñ™Mò—9Þ ›m}xé È8ë[f¹S‚àî[c‘a}©íÞÊàO—gÿ0M endstream endobj 929 0 obj << /Type /Page /Contents 930 0 R /Resources 928 0 R /MediaBox [0 0 612 792] /Parent 954 0 R >> endobj 931 0 obj << /D [929 0 R /XYZ 71 791.866 null] >> endobj 932 0 obj << /D [929 0 R /XYZ 83.504 755.001 null] >> endobj 933 0 obj << /D [929 0 R /XYZ 83.504 743.045 null] >> endobj 934 0 obj << /D [929 0 R /XYZ 83.504 731.09 null] >> endobj 935 0 obj << /D [929 0 R /XYZ 83.504 719.135 null] >> endobj 936 0 obj << /D [929 0 R /XYZ 83.504 707.18 null] >> endobj 937 0 obj << /D [929 0 R /XYZ 83.504 695.225 null] >> endobj 938 0 obj << /D [929 0 R /XYZ 83.504 683.27 null] >> endobj 939 0 obj << /D [929 0 R /XYZ 83.504 671.314 null] >> endobj 940 0 obj << /D [929 0 R /XYZ 83.504 659.359 null] >> endobj 941 0 obj << /D [929 0 R /XYZ 83.504 647.404 null] >> endobj 942 0 obj << /D [929 0 R /XYZ 83.504 635.449 null] >> endobj 943 0 obj << /D [929 0 R /XYZ 83.504 623.494 null] >> endobj 944 0 obj << /D [929 0 R /XYZ 83.504 611.539 null] >> endobj 945 0 obj << /D [929 0 R /XYZ 83.504 599.583 null] >> endobj 946 0 obj << /D [929 0 R /XYZ 83.504 587.628 null] >> endobj 947 0 obj << /D [929 0 R /XYZ 83.504 575.673 null] >> endobj 948 0 obj << /D [929 0 R /XYZ 83.504 563.718 null] >> endobj 949 0 obj << /D [929 0 R /XYZ 83.504 551.763 null] >> endobj 950 0 obj << /D [929 0 R /XYZ 83.504 539.808 null] >> endobj 951 0 obj << /D [929 0 R /XYZ 83.504 527.852 null] >> endobj 952 0 obj << /D [929 0 R /XYZ 83.504 515.897 null] >> endobj 953 0 obj << /D [929 0 R /XYZ 83.504 503.942 null] >> endobj 928 0 obj << /Font << /F57 146 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 959 0 obj << /Length 4421 /Filter /FlateDecode >> stream xÚíksÛ6ò»…®37£ÜÔ @ô&3—kÒNúÈ´‰ï®3i>ÐmëªW(ÉŽï×€Ô’¢lJ^Z¢›/0Møï],¤wÙ#½ïOˆÿùϳ“çß Ý£aÀCÉzg=Åz*ä EïlØûЗÏN)!¤ÿíì:NGñt<;e*ê·š–£ÙtñìãÙfÕ£$ÐDÓlÒ;å$à$rc¼|vÊ•îÀÜ qá‡p#(G DZR3’áׇrÞSVÄö‘‘¢wJƒ(d®Ûï$$~4]Í|£=Ðó×-*MªFy>åo~ ^^–™]æ]~'„ÕCi ÍDwÄ©TÞå/® /w2ß综«ƒ³…½^Ž 7£å•ƒöÕíÔÑHÎ+$©*ß&"Úˆawâ‚ÖAº„‹¯kyÈ@šÞ»^¶Ò‘0ö1C™®G£fqœ^×£Œ©beÖëWUñ¼–½È`¸mFÈðþiÝ¢NC!IC'‡¼ÒœÔjšæ-,Î/°¿ lFI(N["÷½È§†ßÙðRe,“|Ò­`e‡[{ýµúÙVèœ]4Fφ¡ÅêÏ“åM’LŒtzÃ<Œ¦F›,ê>]öcêNY¯¢ O‡uÃ0b¡÷ayª*ƒõ\Ôü¿°6¨Ñ„:P‘6EÀ"å-Ñ€z[ôÍÔX4q®#7lO&ƒPxCïå¥Ññ_‚h-N¸V>€ù•9›Ã¼^Œ&óq’?Oó?Ϧã[÷2¹ŽÇ«ØÚ öwk™ÅÀ«ñØ=•Œ]óû$^¦£Ïuå 蚥¬5`iÀE¡m_ÔÒ “M fz·ÁÙ•:6W¿ÕFÑ}:PN@ thrmd,Ïl½°j륣kƒŒë¤–îEFd¼ Õ°lÂ}bÀ‘ íŒÏúà$ì^/jÆ2vIdzphïôü#ëuòúìäÓIÆÝWÆ¢Æ ¤=i>ƒšÑ““IohþøC¡1Îol×Ij'Ï{ïO~uŽZyÂ|,"ðö,ò€Ï°'LgóNÆDÆ3ŸÊ#]úÔ*6Có ÌâX l©f$ /$Jœ1—ý›Ä’>Å›A:[,Ü+Ãn“Z,Q³|¥¢ªE«ê¨VòBî^¯ Xæ߬;¾Ø"p9ã÷2ŠlÌ(¢)£„›E…QÔ½7½[7•Á±¤½RÇmœ©I¸¡Ê5 Õî8ܶ8ÑÖÍÉŠÈ]‘P†íŽ ë…<¾˜¥Ž¡çi2<$" "*‹å8õ&Iš|ZÌX_„œòŸvÆ,° C«ò¹*ÊøT[sÅùÔ².tâ{úÅ]¹.ÙgmÁ‘ÑÄE°*˜_»ž[í£ ÍY?©±2 ýN˜…šì¤÷G/úÑ}X¼bVôžjD,£Ak߸h$è´—f¿ºÇw@ŒcÖ¢Ö¾Í_ p÷_{NeoxvFý™m¯m›VõSû<*ÞDý©}ïú'(Ó+c/ÙAW¶Ú¡öyiÛ‘mg`úÊÄ’dµ4Cάë܇Á2€pm>ÖpÖ Ô®Qöoq†ÓýïÍꤧ–•GNöfP{Ä¢P éÏí©mg¶Í¨%ƒ{>±D¢EiBÚ‡¤éøFí@šfhö,ª;3‡FfmbÛ 6‰d$µAXmE£CiYO÷oì el"N½,‰Sv4âtˆÿÂ’ˆ#£ÕÍSÚX?âG–¦¶Úv…µ™•pXò·ªw”×2U½ÕšÞa7+€ 7AUïäÓ7enî6–ÎÜNú »ÑŸØvnÛ±y(ˆ`žÅ§–‡8ƒ†^R:N,çàÍHÐ¥g¦²óÁqƾËCn=±)Èõ X›æz§Ò#D®ß™¼7&O÷~,ˆü-½ôþ¶¼Ü\¶¿".Ìl§8œÀ¶8DŽDÕ†C”û+Hâ8S°™8žYqœâ ÊK⸌œ\ñJT{Ä[Ïû+Ø}ªìqï·Ìl{mÛŸ,ÞØç·í¯"”:ÂbâÕ„SŽÀûÛ6}<õ¡6xM}Ð`…£i¹AÓ²âñîKͪÔ,KòÅ©ƒŸ,äßØç·ó^!Ø™÷®‹iãµFߢmú>@0Žú–["²Ù(šIÚ(³C3; sñ\ŒøÔK«IßYMú …£§Ä0ªš|áÇ/üØ*?J–…Ÿ‹¡cÛâØ|º\©5iOM?Ûç¸0Œµ%c§%µ·aFöyáÿ·ý•2ŽÅ§f Ÿ‹=a?0çÓÔsîS° ŸÊ=ì½™Iµ§7mô˜ ó5MC/²äì~–õŸÛgöìT ·—ýð‰>·Šü¹mÖ› g§XÉ|F9±ÝÅš´‰ ¥X%{È5 " z[±3äãÙ˜Á R²-V@rÅ>8ŠÂÍp™Ðpz8ßãtû`y±GlÔÈJ`èA|”§ZЮóÑ ˜{‹.ò÷Tå”ï°ØÕv“!òFOõ¯îÆ Iò?›• Ö'îï-˜KmГ7Ê_pc&¨c6ʉ¸9n;|éó#ö6Ýîú L°hÍ|#¦´l¸t†è²ˆîʧw¶ý$M=Z:¦|¢~ÙñVæX šhøx3.\<µ±”×ea;º,—IGSQ·Í!Ÿ†<è´½²¶Er#a ˜sø‘™3gÈîD KQ ˆrIPDå$Õ¦#¢v4ŽZL^”[NZJ,GDu—±ßÛöu÷½3?S0æŸÛpbQ˜‹å~.¤…9fåMž%ˆ÷¶æîì„I“MžŽ©{Ӿ‘ ÑÓ _Ôý>â@âH—bõâH·i}|*ɺK’åé÷ô¯ÝCùW§R¤9±çzÞñïØîqŸƒ½ïkÜRçVˬ¬k¹ÄÊ9¾)è§ÿkB¡ŒuÝ7”v:™ñír·¥,ïµ*be|_'Õ«FÀ¿ê_„lËR’=Ü¿¾ÆM"éèÉ¥½I[v×Púlk,n>¦¡Ä•$å%²ö¬ ¯nVè†åÔ«Ò‰¡I¥Îj¢tQ;õ¤1UZ+qcŽb—þ…_ÕcÆy%:’˜1ñ¬?^þ 8Ý"qs+óƒ`+à¢ÁTß…_R¥Ð2žóý—|ô|0iãüévÂÔS©ÛŠ‹GíÙŒº»Šõ»¢>¢´uÏeÿÛn)Öã¾s ´ÄȽ†vÛHC¬¡Gv¥S,+b°‘ÍñßÂË« *îFGÔÞyË\I¸äEkÄïköPp–ø鸸’ ì_bf ˆMðÐNÉ‘6Áy'jïPâ¡››º—@ƒ%Àì]Tn¬ÀâÅÀ¹¾:/íþä{æj‡ZÚǪ:R”¶7í-å2ÁÔWÖ'Å—!y¨š$jdõ ±œ·ö[¾§á{öH¶y(x€[2WÐ3E¢ wÏÌæå5‘…2³PfÊ Êy&ü¼HXÕ¾¤«ùälLŠÌy]J^ùº‘HvÖ1…Ð¥†øïËÛqéÀßZ–žUXGˆ‚óãQ2y r&õt/Û¨Í9[—ÕìÝ ”º al-Œ­¸cªë#º^ïq²ádØÂÅa’øÊ[ÎÜŸ—žq$[Â᫆ |6šç¼&žÕòmú_ÏwßÀïº1/üÍ ì q¥¢†©nFMÕxäF)D/±ÂxKàÄ`n |ê#=‡üædbå·IZ’jú__aÅ©&à~Â\P‡&lFEäŠ[–á>:oæÅýx¼Ñy&kh3¤"Æ;`Ëiä,“akøä°ÙÍÿ<ô½¬¤Ö·x›ùãúIɶ,{ñåMh‰†¯À%¥K𜶗,݃žë„PÑÔ(¢T!Ñ­³Š\ä jÜ[„ŠŠÐ%%OŽ)y·ÜpÉAâÍö»ÌP¹˜x>M6ò;V¥#¬HaT×›úpB ‚ q‘(¡}-æ%x†YÿÞ1Oû ¤ ,×H)é>T^u„óCXè•Úç"ÃìÝ/KJ‡Éœ»¸ßÝÒ{Ëhy§¹:’p Ï‚}Øbìè»etWx€ƒmQ:ò»‚{ŠeÓÅð€hZd‡Fw¸Ÿ‚`&àIN½%ãI¶‘ñT—p Œûj‘…»÷¯K5æ“ÂÕÄ:‚Í££z-ØnÄ›±Ÿ1;ðuú‘<Ìs@ W@»^WÂ̹èK±Èr¨^’:y²8Úá¢ÈgœN3N±QMý‡¯ÀÃy©à7ÊÚnò­ô·:8knÛ•*!¥HL0°bí¤ÎÀ-EKp…û ³ü„.­— ”3æ*ÜÕYàì›é£Ù7›ƒ}³h1O …ªçÛao¾÷ Ú»²,Ëæ nFþ-Ø‘YafÊä—)fØE`ÓOì#€x"KÈ"ž–µ7ø ZÐVU8…1õÛ ë-‰Ø’˜6>T‰ºaæ¼µhœùŠK6kHŽF>/ü-;ëC‡+PÇaˆuÞèL0­{ƒFÍ00;ª¹5÷_0Ø8 …•™ÑÈ ½|Ö­ÉgIKˆMü¸cý¡‚;´—bždíÕ.Ú¨î}›ËHðÒ"sÞ8§'Iù5ɳJ°%¬ ¶<°dn0Ò í0ÍʳR@\¼(Ú¨t\ÙC­4ìjÞ"ê"äŠ"ò¡Ç\›5ÁHçª\"î>ùŒÄAçß}ƒ#ŸØqlP…G’ÏkЕ¢Ã;#öÿÓ–vþ’ùV>áÅ‘¯~§ÞäÜ55ïêGÕv¼ŠóËM j̯ùR|—»‰;É’n«É˜ÿ´> endobj 955 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 576.188 51.396 584.628] /A << /S /GoTo /D (noweb.NW2i2LRH-4Mxm99-1) >> >> endobj 956 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [152.844 573.551 164.807 586.447] /A << /S /GoTo /D (noweb.NW2i2LRH-4Mxm99-1) >> >> endobj 960 0 obj << /D [958 0 R /XYZ 125.795 791.866 null] >> endobj 122 0 obj << /D [958 0 R /XYZ 72 754.004 null] >> endobj 126 0 obj << /D [958 0 R /XYZ 72 667.24 null] >> endobj 361 0 obj << /D [958 0 R /XYZ 81.963 577.274 null] >> endobj 961 0 obj << /D [958 0 R /XYZ 72 574.547 null] >> endobj 962 0 obj << /D [958 0 R /XYZ 83.504 577.274 null] >> endobj 963 0 obj << /D [958 0 R /XYZ 83.504 565.319 null] >> endobj 964 0 obj << /D [958 0 R /XYZ 83.504 553.364 null] >> endobj 965 0 obj << /D [958 0 R /XYZ 83.504 541.409 null] >> endobj 966 0 obj << /D [958 0 R /XYZ 83.504 529.454 null] >> endobj 967 0 obj << /D [958 0 R /XYZ 83.504 517.498 null] >> endobj 968 0 obj << /D [958 0 R /XYZ 83.504 505.543 null] >> endobj 969 0 obj << /D [958 0 R /XYZ 83.504 493.588 null] >> endobj 970 0 obj << /D [958 0 R /XYZ 83.504 481.633 null] >> endobj 971 0 obj << /D [958 0 R /XYZ 83.504 469.678 null] >> endobj 972 0 obj << /D [958 0 R /XYZ 83.504 457.723 null] >> endobj 973 0 obj << /D [958 0 R /XYZ 83.504 445.767 null] >> endobj 974 0 obj << /D [958 0 R /XYZ 83.504 433.812 null] >> endobj 975 0 obj << /D [958 0 R /XYZ 83.504 421.857 null] >> endobj 976 0 obj << /D [958 0 R /XYZ 83.504 409.902 null] >> endobj 977 0 obj << /D [958 0 R /XYZ 83.504 397.947 null] >> endobj 978 0 obj << /D [958 0 R /XYZ 83.504 385.992 null] >> endobj 979 0 obj << /D [958 0 R /XYZ 83.504 374.036 null] >> endobj 980 0 obj << /D [958 0 R /XYZ 83.504 362.081 null] >> endobj 981 0 obj << /D [958 0 R /XYZ 83.504 350.126 null] >> endobj 982 0 obj << /D [958 0 R /XYZ 83.504 338.171 null] >> endobj 983 0 obj << /D [958 0 R /XYZ 83.504 326.216 null] >> endobj 984 0 obj << /D [958 0 R /XYZ 83.504 314.261 null] >> endobj 985 0 obj << /D [958 0 R /XYZ 83.504 302.305 null] >> endobj 986 0 obj << /D [958 0 R /XYZ 83.504 290.35 null] >> endobj 987 0 obj << /D [958 0 R /XYZ 83.504 278.395 null] >> endobj 988 0 obj << /D [958 0 R /XYZ 83.504 266.44 null] >> endobj 989 0 obj << /D [958 0 R /XYZ 83.504 254.485 null] >> endobj 990 0 obj << /D [958 0 R /XYZ 83.504 242.53 null] >> endobj 991 0 obj << /D [958 0 R /XYZ 83.504 230.574 null] >> endobj 992 0 obj << /D [958 0 R /XYZ 83.504 218.619 null] >> endobj 993 0 obj << /D [958 0 R /XYZ 83.504 206.664 null] >> endobj 994 0 obj << /D [958 0 R /XYZ 83.504 194.709 null] >> endobj 995 0 obj << /D [958 0 R /XYZ 83.504 182.754 null] >> endobj 996 0 obj << /D [958 0 R /XYZ 83.504 170.799 null] >> endobj 997 0 obj << /D [958 0 R /XYZ 83.504 158.843 null] >> endobj 998 0 obj << /D [958 0 R /XYZ 83.504 146.888 null] >> endobj 999 0 obj << /D [958 0 R /XYZ 83.504 134.933 null] >> endobj 1000 0 obj << /D [958 0 R /XYZ 83.504 122.978 null] >> endobj 1001 0 obj << /D [958 0 R /XYZ 83.504 111.023 null] >> endobj 1002 0 obj << /D [958 0 R /XYZ 83.504 99.068 null] >> endobj 1003 0 obj << /D [958 0 R /XYZ 83.504 87.112 null] >> endobj 1004 0 obj << /D [958 0 R /XYZ 83.504 75.157 null] >> endobj 957 0 obj << /Font << /F59 148 0 R /F57 146 0 R /F70 201 0 R /F133 581 0 R /F79 203 0 R /F80 204 0 R /F73 223 0 R /F33 580 0 R /F136 582 0 R /F77 565 0 R /F74 202 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 1007 0 obj << /Length 3359 /Filter /FlateDecode >> stream xÚí][Û6~ϯ0 ,¢ÁÆ^Io,Ðn“ÝA»m:À>dóàØžÄÈÜâË\Zä¿—:$eÚžLhû“-O¬£±)Š:÷ïœÃ´õ¾•¶~x’ºÿgf,ÿûØòŸŽŸüûäÉ?^qÙÒ-rÑ:9m Ù)Š¢% ÖI Þ:é·Þ$Eqôöä?ÕDi«Ã[¬5*ÿi?¾þ¡œ‡ód¢ÃòÜ|føÛÜ÷­dam÷ûv–u4÷ëàØuÜ?Á&m½içLäÉxãû˜™K“A&*tÒ;j"¹¤ñšÆŸiüŽÆŸh|‰yK¢!Ô"wO-´ù,99j3|0›œ&Ȥ¢ 7ªè*z£*éšQ&#ú<¬®¨ä‚®Ûïcn/óä”&áÌn “ Êܤù+ä–)=™0OPŽííGÕ"ˆó ÃtQͺ8­—ÓJ9m§õòä=î6ˆ‰xêvÖîò­ñŒ>ƒ]ƬZ0Ç—Ä7˜'Ðɹ™4'ÚÏi½yr‡Y¯*×k¶èÎLª“+te\éÒ8¢ëöó¹ÿ¦'t}à¿{sFë˜7ÄJJƒHOÕéYÐ^eÉ-W4Ú+='4Æì¡™.KUò ³.+û bƒ±(æ G«WmX¹}탿¢>û&äöV=äôLžC'˜÷!’oâh€çÉ‹Èo²¨9# 9íx$¤-Õ§å62#~e9HY’ˆB:¢WD!¡Daö%mÎ)fº¼ŽÍIéÁ ÇáW4ZþǯിÌMÝ¥±Gã ŽX}ÔÁH~Ý»ù½ÂKz# uÈ9uÒ¥ñbçüµgÛóŒ­Zƒ:ÎôPï[ÉÓ£v.“ÿ§Rƒ›K¬5q¿ÚZ[’7%ŸH{y ãw1ï¹à 5˜â 9lÅäpÝŒ ’Hر½ì,Ô]yî£Ê¶c¸HÍ:t>ú} cvÞ¹˜€ÈÂ!ù^€w2KŽAÊíëþQž‰f"|k "»Bøv€ˆ!{M#öÆ0ÔDì.yò‘lù1¹Dc”›P~Èw‡‡o?$<¼‹h5»±JžF'C÷‰»¬“a#ïc#N6Œ®!0¥8À”{S g>h«8IZf¡—+~í_ÎÐ÷†3lô.NÄðä9Š2ñ:‡gÐ`Öû]“òÐØ ">—.îŠõ@#¥ÜQ½å€3OQ%?="LJˆU9Hðkĺ¶n{Š/i5„¹mÒn-2Q0ÎîRHa@GcÔ† 8¨¡Â[¤¨Q+˜¥"Çò¹t½ƒ¹ 䙌ú€Bë¥Ö«›ÏQ以,Æ›àÊéSù¦£ fo¸t³ÏU½®5u:B*y³Ò[ Â§ˆmË|*\Õ0Ø|Z-ŒŠ …jHHG74,ö4´È^5ã”ÆŠâgp!«.TA­ŸtÕA‹ÉO6ðDa Z?Ù X™íΡŽï“P?&¡>Üh(µÚf¯ùË•Qñú†žõí‘ùnÒa´ÿƒDd¾¼É‰7É¥®l£mphž³C¡ãö84GÝ’`ý<>EGá@Ùmr±s“%Û.–ÄÅE‰íTQ\ß äà˜u_³ ttVAdˆÕ² ´(“ x¾Ÿìî•vCòä!áPný8Ë­mÈÀ·VÏ1œk¹zíýª–uâÍ“ Ð| .çŠ ¯‚ò^(Œ™Ô«˜˜îtSWPÂÂæ”Ç|ƒJæ \n²Ïg혎µÕ̯–WëzTþdëÖMU>hŸg3ª|ëkÄ»¶Dj >9ßèÞ6ˆÜ&šà+F—:L,ôŠˆl»ü ¤êmÍ”F ù¿ ÒkF¨l‚x·ÄÉ]q>¢c#|¿KaÙ{Q_ð^üíÇÀnñLWnHDØDŸ0ÃÚ<øUàOõ‚ö§ÁY#ñ#³ «¶å7.A´Æ› Óþ ¸ÇY6jºÜ@€ ‚ä'Ÿhˆ–“Üw¿i^Fø*'À¬+³UÚH*°þ ƒ®eÞÒc›=ûìPZ?ì>køáSäÖsCäå.e´KÏQ9h6þxÊ_‚qC{4¶¨éSt/ŒÏÀ„-ö…âû+R<;P|ŒJ3Ò«¤3ïbýIµ‚?éOˆí̲̄Ðò`í4#lñ}~AŸ_¬dçHë­ðr®ß–(•[u=§ñ­¿"ã=Y‡•R£í®ÔŒ¡æX#öÛ ÊÉ+¤ñ„ÆïãÍ„g[~Ž¹œ8äÄ)"vIÄîÑæ±ûN,¾ó[e>ÝHMÔÔùTgó†ØUo€äN±íˆGÈ$QÏ1ÑÃàÀÞ~ÐÛ¦·ÐaÖUZÏUÎ]å××q"¥wpzižu”Èæב¥k€¦n"tQß8h`´˜l#]²ÍÞÒÈܸ[o›t[Ynú.±5l7 Ð6ã"ß,¤JúÎá°Ç¬Œ±Ð°‚o¼"ÚµRhÖŽrZYpÜn;|1CF2:ÈQA†Œ^ÊQ[Ï1oîäžö$ë Ô¬ÕðúiàŸ\DÃѵ«DòL£ÚÄWQl§¾A­4ÌÓ»&Î ½Ø2Æ•E.µŒqÝvôbËž.´Œ‰M‹Zœï[º¨ÀÓÖâ³CkTsb©T»5'Hh3§?§ q°Jë‹H„¯zL§æ%±Ýš¬“µØÃÃþŸ„e–vtª3+ãd§(ŠV›i)™uù|èíåÉ“?1//Ç endstream endobj 1006 0 obj << /Type /Page /Contents 1007 0 R /Resources 1005 0 R /MediaBox [0 0 612 792] /Parent 954 0 R >> endobj 1008 0 obj << /D [1006 0 R /XYZ 71 791.866 null] >> endobj 1009 0 obj << /D [1006 0 R /XYZ 83.504 755.001 null] >> endobj 1010 0 obj << /D [1006 0 R /XYZ 83.504 743.045 null] >> endobj 1011 0 obj << /D [1006 0 R /XYZ 83.504 731.09 null] >> endobj 1012 0 obj << /D [1006 0 R /XYZ 83.504 719.135 null] >> endobj 1013 0 obj << /D [1006 0 R /XYZ 83.504 707.18 null] >> endobj 1014 0 obj << /D [1006 0 R /XYZ 83.504 695.225 null] >> endobj 1015 0 obj << /D [1006 0 R /XYZ 83.504 683.27 null] >> endobj 1016 0 obj << /D [1006 0 R /XYZ 83.504 671.314 null] >> endobj 1017 0 obj << /D [1006 0 R /XYZ 83.504 659.359 null] >> endobj 1018 0 obj << /D [1006 0 R /XYZ 83.504 647.404 null] >> endobj 1019 0 obj << /D [1006 0 R /XYZ 83.504 635.449 null] >> endobj 1020 0 obj << /D [1006 0 R /XYZ 83.504 623.494 null] >> endobj 1021 0 obj << /D [1006 0 R /XYZ 83.504 611.539 null] >> endobj 1022 0 obj << /D [1006 0 R /XYZ 83.504 599.583 null] >> endobj 1023 0 obj << /D [1006 0 R /XYZ 83.504 587.628 null] >> endobj 1024 0 obj << /D [1006 0 R /XYZ 83.504 575.673 null] >> endobj 1025 0 obj << /D [1006 0 R /XYZ 83.504 563.718 null] >> endobj 1026 0 obj << /D [1006 0 R /XYZ 83.504 551.763 null] >> endobj 1027 0 obj << /D [1006 0 R /XYZ 83.504 539.808 null] >> endobj 1028 0 obj << /D [1006 0 R /XYZ 83.504 527.852 null] >> endobj 1029 0 obj << /D [1006 0 R /XYZ 83.504 515.897 null] >> endobj 1030 0 obj << /D [1006 0 R /XYZ 83.504 503.942 null] >> endobj 1031 0 obj << /D [1006 0 R /XYZ 83.504 491.987 null] >> endobj 1032 0 obj << /D [1006 0 R /XYZ 83.504 480.032 null] >> endobj 1033 0 obj << /D [1006 0 R /XYZ 83.504 468.077 null] >> endobj 1034 0 obj << /D [1006 0 R /XYZ 83.504 456.121 null] >> endobj 1035 0 obj << /D [1006 0 R /XYZ 83.504 444.166 null] >> endobj 1036 0 obj << /D [1006 0 R /XYZ 83.504 432.211 null] >> endobj 1037 0 obj << /D [1006 0 R /XYZ 83.504 420.256 null] >> endobj 1038 0 obj << /D [1006 0 R /XYZ 83.504 408.301 null] >> endobj 1039 0 obj << /D [1006 0 R /XYZ 83.504 396.346 null] >> endobj 1040 0 obj << /D [1006 0 R /XYZ 83.504 384.39 null] >> endobj 1041 0 obj << /D [1006 0 R /XYZ 83.504 372.435 null] >> endobj 1042 0 obj << /D [1006 0 R /XYZ 83.504 360.48 null] >> endobj 1043 0 obj << /D [1006 0 R /XYZ 83.504 348.525 null] >> endobj 1044 0 obj << /D [1006 0 R /XYZ 83.504 336.57 null] >> endobj 1045 0 obj << /D [1006 0 R /XYZ 83.504 324.615 null] >> endobj 1046 0 obj << /D [1006 0 R /XYZ 83.504 312.659 null] >> endobj 1047 0 obj << /D [1006 0 R /XYZ 83.504 300.704 null] >> endobj 1048 0 obj << /D [1006 0 R /XYZ 83.504 288.749 null] >> endobj 1049 0 obj << /D [1006 0 R /XYZ 83.504 276.794 null] >> endobj 1050 0 obj << /D [1006 0 R /XYZ 83.504 264.839 null] >> endobj 1051 0 obj << /D [1006 0 R /XYZ 83.504 252.884 null] >> endobj 1052 0 obj << /D [1006 0 R /XYZ 83.504 240.928 null] >> endobj 1053 0 obj << /D [1006 0 R /XYZ 83.504 228.973 null] >> endobj 1054 0 obj << /D [1006 0 R /XYZ 83.504 217.018 null] >> endobj 1055 0 obj << /D [1006 0 R /XYZ 83.504 205.063 null] >> endobj 1056 0 obj << /D [1006 0 R /XYZ 83.504 193.108 null] >> endobj 1057 0 obj << /D [1006 0 R /XYZ 83.504 181.153 null] >> endobj 1058 0 obj << /D [1006 0 R /XYZ 83.504 169.197 null] >> endobj 1059 0 obj << /D [1006 0 R /XYZ 83.504 157.242 null] >> endobj 1060 0 obj << /D [1006 0 R /XYZ 83.504 145.287 null] >> endobj 1061 0 obj << /D [1006 0 R /XYZ 83.504 133.332 null] >> endobj 1062 0 obj << /D [1006 0 R /XYZ 83.504 121.377 null] >> endobj 1063 0 obj << /D [1006 0 R /XYZ 83.504 109.422 null] >> endobj 1064 0 obj << /D [1006 0 R /XYZ 83.504 97.466 null] >> endobj 1065 0 obj << /D [1006 0 R /XYZ 83.504 85.511 null] >> endobj 1066 0 obj << /D [1006 0 R /XYZ 83.504 73.556 null] >> endobj 1005 0 obj << /Font << /F57 146 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 1070 0 obj << /Length 6645 /Filter /FlateDecode >> stream xÚí]ë“Û6’ÿî¿Bùp¹bÑxÌž·n7ŽSÞ»¤Û[uuvª"k8c&zL$ÍŒ}ÿý H‚Tó!Ší”ËG‚ÐF÷Ý™\LÈä‡{Ä¿Sxµÿ~ŸäOÿ~ïï/ï=|"õ$ŽbÅÔäåùDÑ(&r¢ˆ“—g“WSJØý_^þ£¨‰L"9‰èdkÿÌŸÿ`+’AEŒFFYJ®Š©T€±Rãcæ?£À,áã2‚WpÊ·“W3j›žßŸ 1MÜë5¼òéÜ=/{Ї:4™¾&’ô+¬ÔtqÆãéH ×ëW†Nÿʳé·îõÿÜ+¥#u•¸+:#?¾ÎØNzzãTà­{MÝëÂ2¥§[¨N9­SÓ½{½r¯Ùçkx•ÓÝ8ĸ#@lk@ó’q*eÀåqï^·î5u¯ï£‰ÙôkàZN9¼ªé7îùs£Óǽ ó'ê^qY™ê8E¹ø „Ú¯$›F½ëLF­³W›ÇWAÉzõˆmG쬗9ÔeäÙô¹ƒúŸÝsê^w¾L¯:)%#)´Ãr— Æ™OìÄeœYÅŽ;3}㞯Üó>ø|ãž×cáÒU ŠÄ‰h7¸«»îœLùX£Ä¥2wa[<óé*p)öÁóh‹†‰éƒÏÖü^8Q|ïžçîuë^ÏnÛüüÈê<iä…a°ïðÃüØFÇéêdJ,˜@‡ŸÝr‡+‚tøOÁ'iEà·Üáóqº7ö>PÕﱯK÷:Ž¡A€aqF».ÍÇìÙ8œÑNeµ1_¦_\ᣫxÑH(¯?r,¡rF̉%¸…:‡ºÖ GÂÞØcæòHÇ¥GÕàÂrºq\/z»Ï{†ËÜÁ­vL+¯Ó…«éwîyã^¯{×õöhW½½ä±L"¾#á5­Ÿh^%)ã8'éâ܇O Ž²ŒVy‹Ë(ÓfºØ\¾ß¦o÷ÙO€?QÂå¨LDb‘ÿFγ"ºq$TA¶ìäœÊx’5…¹æÅ,ŠŸÌ8°›ì§*b÷g”2}ºº\&«d½OÎ2¿Û\Ï·é|½H²¿Ÿ\­ût³ÞSÅ$¦¶rÒS‘4«öEºJ—óíò=ü4?jcßÙtÿ6É>X%óuötž×úÀþ­¦7¾Äåvsž%ÙïæÙg7o7Kÿõ|y‘¼Ùγ¯7çÙ‡‹cøü¼Ê±&!Ç‚À³’¹ ÷eâ°LñRŸ¾ÍJ˜J-2R’ç%þ³è¨ f.oæïw-–5G…e…‘fÝF³êôªÒüK‹ÞMÕëIzÇ tEï8‹´òáÊO›t—´(Þ0² ŠžŸ¥ûôÚ; †»«­‹–¼WŒóU¨ß0¾€=äë5HÛ£¥¨ ª3ÒRxÓ{éå\J½fŠa.9\—U>KˆHUŸÑ„ׇ‚J ‰"®aUÀ£Òô '¶<†b¶vàr:¡ðH‘÷ïäX8šÔ¤%§Á¨ZªQ`šÈ°¬áß+¿oA…aTP! [+’?® Rô#Rzkóõ¾†q×€ wpëCÀ…uM¸%ã8š}T±e{ÝÆAÀ‚ìf#fÆAÀåö~ê<ý©Å4†Ñl0€ê2]'óí}JiTó]œF’é®Ä™ gŒ¡)/¨¨T†çˆ&ØœQ [lq 8bÔqàÔ ¿J¾þŠv:RàõÃTï·Ôa lðIƒ¶¹¤GÓDMCªÓÌK€„¸Û3VZu™—ˆ‰ŒÜ¾æÛ³fH7± eocnœ)¸géüb³žûi†›ÄÎ1¤ë Ô ©ˆ#-ÌXfhF2á"CÍ°"²#ì°–80Äׄøql ¬-ŽXÅgEʤb¯¦(Ãýä˜Åã™ý@V•%7ûŠ@ϱŒ¾ðÐûáL"r’PÆNTDÆgÊ)F dªX”ypF•žF‘{P ôÄðCú•µôyÅ‹xŒY„,Öy*K]4ƒáñ=™ˆO‹Ï…fOFð5„VÞs0ܬ[ôtÀ0¤Œƒá›t¾C‘Oƒ“IÍ]ó?ŠGãÀŠç˜ôîmz)‡Y ü5}ƒ™|5@›’Ûa:‚¢sÀaëýŠFB0B’"#Ñ¢ÜÊV”»†þÚ:G3††l¯¢t¤‰Ó£x`v;öS}Ï6Ë÷-3Œjä„t/îz³JçËú4#îlQè!ïæ “ž¨È§ŒÌkÞ 8c÷d2Ó«z:>óyÄdëà¦ú€› Ü$nf€Û¯ˆSŽ€[(Ĭñ%­³fܦ Þa¨ãx‡ ¬+VÎ::Ÿ¯¡2é²­dHV̶¤X›Ÿw´LQ¸ì´ItpLÁÈE UtB®0ÓÞÉ~ñ}{Ì;Œl憄ϯ–mw\Í·ùZµäÝ%øë=`1Ž½ŒDDð;‡½ÃäÔ€½¡œNÅÞ6V.gèK|)¾:­ˆÊD< 㢨¹ùs#ÈØ_eýSk¬[LÆrÕ`š'%¢Gø]AÙÛÉêôJëÜf^§¶ª°ž×év3ôXníëg­¨@†.îÌë`q$ëdƒIÞégØåé Hïg¤»–…FÉ6ø!ájý8_ä“»ç “Uƒ¯Êêsô5À5g`„= 0F¥Û| ý‰‰û8¾—ef½VÉùØŒ•±˜fïp:6rÂ@d‡ËzêË2!.Îý2Üwx-ŸnóýÑ߆ ~Dˆ ãø¿âîãGÄÿ·?0-TtP功60 Ãx'\ÐÎÙíº‘ÁÀüÏ–‘y݆‘9¤|ÜXL(˜Â[g2ŠtŠ±8”N1â7ø‚Dw¯q{·§†·‡ÐÓnäk߬.ÇΦúïkBy“7M¿ðhßõa¦@¤@krôa<Œå¢Â}Í«Íæ¢ãê@jˆBú3—z»¹ÔzËk!ÛQts¨«­‹¨.æ“[BÀãu ug”:m(7öÀÑɆŽ;]"nlar—¨57;,îU.Ñ0Ùà.QE6£®¹iôˆøíxDP »õˆâNˆ›8Ò†µzDv›ìãùîéXhª=ñ"JÿÌÄ6fbÕ§çÏœ”‰EÜÕÇaýÜ™aƒ»3€ùSº¼—Àoß?8º“б]èÓæn¹HïÚl×Oéι[®ì²za?ÎíñPí>Â0Ò >Bøk.évz‚º°þ¶¶=™ºW€yÇ•lâLû*絜#¦ÓÇöËhdoQY¨Û–Åñ{¼&Y¢mí8–ƒ^²XRÈ“¢ ߢ\ÖÚ«¼1:î×>þ¡Ú×06™ºÃƒ,ýe5žê€ð4’šÈ(Åš*RlbžŽ©Ç«8ãE»*¹x{d¢Â’´>¡‰Ã‰ŠšOk=_#'\SðŒD›Ok ³Y›K è ð£ê…Ö— Ãㆸ$ÄãqSµõ^Ì6Z!©M‘8Ãw V9ö âlQzº–éK6±U8XÇÎñ€là¬5ñÎÆã½"+± ƒKÆOfÁ(z„®‚±ß.ãgn»“+÷Ç¡EH ¾™s°Àm3ÁØÎ'ÖíÕ™5`…¢7nÇð½a¦ÙƒÝÚJÒW¥„û]yÓŽÅBܹCÐTMïPŒF%ë“t.ŠµD©<¬®=L «ûte€ŽÃF€†À1¾àµ@Ç  Òȉ§/q.»æ׉MúSÓ¹é› ›÷÷g²µoùHµ!Æ è®“«mž^'{4Гà|Ì-—ãzni˜ ŠB‡›¤;Úà$t5tGæ-o8ésGƒ=LÛ<è@Ûou‘õ.]#ҤܭÞ‘’¦§SiIÎnq¯C—¬ä:úcÇåð’ÁÊP†ìÈðCn­<”!¯!Q¡J[µ_p/¨¢2ý~“*H2þÜ.R¥ýoR=PHãúDhUq¤˜ãÖ-¦y”oïºHu Y<¢ wÞ£:2:†”¿ÔkTGhîx…}MhË%ªÇ“Eu_°ÓnbÒz?zû‘öÂïP¿XÌ—- §’m°Ÿðή^S^PÞ`UÃøi°ª»‘ª²kCÿ™Yêt·Ü®ÆiaWu‘6¤ñXyžÃÊTÒóó<š{Ô´„6§™§ÝÍÅãî#P;8gÂn¶ËO8»ZµØæ š ¦RŸù“Í®.{ †:ˆ9Ãp;…rùÞÈgØ®{‘Ÿ[¤Å‰©¦Wj A[ã!²î¢×Â[ëo¸•ÅV¯±wÙwì ƒº¿¦¯¢_-÷ôMµÄ×(3¦"3†)sPÈü’²gÛMËÙ]CH6XrHtuµÜ§—Ë÷ÃÍx_ FÌã2ùóŸF|[F< L8T¬6>Ž"j}êÄ[žen¦p oÙni£:ßѽû½Åˆ‡Qm°ãn1u¼Y禼ÊÝa&]OË‹mŸb>˜MtšÃ™Ôšß[™I}…ç§ wT»„ó« írH,Õhï *ü¿Çø.º ùK§µmÎÑÅ0@‡Ø¯Ñ‰ë«¨ë§·†žèíßNz œUÚÿf]=V ø0¥lˆ˜B¥lŠ–2ä²/DlÂñûiÃVtW—ÁŽ˜µšA/C-áìûÓ†m1<gßÛñy`÷àvODMEX%O[— ™Î+…£Ó…ÒZ$|ˆð·Ç›Az ÑÉBØ”îÓküâ=øú˜g@ãËÇ+›ÀÜÀ’®Ï’wþçäb›xWq—ìÑ[g…ÛÜÔ†äîÖYŸÓ&Èœ6é; JƒÕöØ*{™rð@oË­ /-IÜÊÒàÛu¿9(t…±ººÅ¼¶EÕ7‹„C4£åqs¦Œ™;†ˆrW”¨œ6¡Ý*nL¥UU1A-à]‚ºÄù!϶!2ú áŠE‚Ô J=Ö¥µµõm1e ºÂB[dÜÉã!«?÷ëõéu^?U4uúS|ù ·KÂ>oÀÅáÞrúÄ;tÕ¶†Ö‡xÚ0!È 2€N¬vDNŠ‰>#¸ªèk KÁþ—N–^Q ` “AÝe¨É Õe JxÂ6ÁÎñ…ô–í¬ÀDÑÎÎ]SÙ©Oȶ©kQzŠøãŸýŠ`Ÿh— ÄPÜÔ »ÉŽZª#1&5÷9•ßÿÖ IçZˆ¼[Öô¯¦¨ƒhWøád5€@‰¨p3wåç3ÛÒàĹƒÐ‘=GÓµñå[ëQZ8xÙ_•lügÓp}îÝÏlBŸ ¢ˆ"|ÞíçÛýÎËk»Yy¹fo›u2;KWÉz—Ÿ¬{¤”µaÈœ/oxë3ºõåçmu¾fBÇγùÈ3Ï|Ùe^A*ú²ƒ/–…1ú|¾ÈžJUôfŠ/%ŠË“¬a=RjL†ñS¢”Ã$kM‚ɈóÎ$X¸ˆîÏ8ãÓ'…eîßÎÝ6&¦7^ÒgÉkBغÑ´¤õ5b‰,2ìHÜ·¤Yz&îo7ÓD»¨ l^õd”j¬1Òn»‚vȈå;å{N³|àpCJ¶ì¬8Žã^p< ËÁÆI€,Gd Nk!†‡c? ï×Éb¿ÙÞ—r:»ž/¯Ü}h±M!]^åó5,²³À%ÌeÆÐ%ˆô¾Zl÷Ø­vžÛóì=¸¦ÍIÿÛ^Ì}‹’s n‹Š¹ISøè[4™þAÀ¯«£“v®ö®áRM¯v™ˆàqïø€‡Ÿ’›ýf=û!ÝÚ»›ÜGç›íêj9O,Úr3}1_].“ÿ®ÀO÷çÙv~³ößd  Oóìí‡gÙûMºë¿(ýSøk±q[ql‚Ʋì­0œfÓùÖ=T\[ø¸à²(¸—*|±¾Z½I¶ÙOœ\á³t½O¶…zÀŽÄÛ$$„àîƶú&˜ê0Ѩçnêh›Zõ‘™¢Dºó­Í_%óÝÕÖ·(oÅÛÍMö`»t™¼Ë:sèôšG’u­]¶ÈÌTzÎ~¹M¢Ò(Œ{öè;—.ãs®2Ið!Óî澪Œ'3}j9xŸß4ø]mÿIu´­Š(A„(¤‡•›•S -Oµ”¾<#©PÄ´ªSA•k•jG$Ša"ñ53WçÖU‘Wá)å<Ô-†ÏîÖƒÉø‡ÅýM“tV’^Õ¿Ò©$å+íëTWýf”ÀÌì-ûïÃU¸úÕÍÎÔ—TÂì4MsÃyׄ¤Zx$uIÅUJ®Vö¿îòX%Ú1ƒQ9 CøºCûŸ^½½–‚&æSó<C9ƒr¥ùô\'k¸Þ@¹‚r€îŒ£ BeÉiŽAg:àÉÊ8içÄDr¥9™%C´P÷¦T~YÜh¡æðts¹¨™‡mîšÆl຀rQ>ÍÍÂ⎻Ь'¶ÍÒ¸4{ì0Øe°Ž¡×׆¨4°N‹RA~sšL¡\C¹ˆCµ9¢Z T›'?ÁõbøŠ´§cW¿vvf1íVÓ YÇš¬Ñfoú‘îóóË cQkà¦Ö¡ Û(WU`i®d¨0YΦA/X~^“ÛM ÊÉ<Ít‰rbP¢\ÕªN¹Ô%Èu8;-]Ÿžl®2¹J¹_e¸l]eû‘|÷C€©‡ VV‚±M0ËPø5S!×ÉSoWܱ ó£X@¹n¦M¬€|˜‚¬˜C:vQjk«¤³‡V ÛJ–%ßÚmh΢à7ö?yò¢öˆáGŸÂ£ƒë³*!›ˆà OpsÝXnòf"ó¯%I¬•À`Ä>"å l}‚ÿ¬ÝÐC¡aðÃ~†0˜m†áƒâaø.µ„S'È8 0A|Ɔ)7þN”n,ìDiHÂÂSb ×o7Q„p~°0vÐúb´Ý«q‚5p俌Ӝa)@`\ئm9H¼j[wÊ·Ú¡Ž;žR±ºg Ž+¤Ž‡šêx°ŽPÇeÊhC!Ù^)Ï4SóYMï ›¾uâý ¼ï&ã¦V^û¹‘äoµ>öÕå~ÒÎîpFÑ璉þNá§n¶šÓÆœ"ø€±bææ°x#Þ ÀßØY3iÔ(+Œ>C¹„Ò¾Ãåa|a½PNš’ÒcŠY¯r0ï,šÈËAæ§0÷¢#Š]áVVÃÊjXY'•4°_¸3@w̵È@„š›Ú[Gp‡¥õºƒ…ôåŽa“;® k‡Ø¹¢ì<ù¡ÅÆö,oh¶¤O’gÔÓÄME æ‡–5¨ÑD*Òdž4‡–4EGÒ4DÞ™4o0iîWtìû¦ÑHž¦lKLÊ-1)Ë ·ÐèJ×åÊ©C¶\Ç™_žüVô×8äKû’ït›|)¯[f·´ÈW&/¡üJ#¤sR‘lNdÖ"+§AVѽr{çåë~2Õ¿Ž¥¥N‘>1ßR‰W±Ô–›‘œËÒ|e Ãõ JgÜnÊkÚiK®»U•9p…î©ïf‡RÙ–¾i§òo ½il‰i°%Á̪œE_-Rý%\ÿRê.XTFttÕZX‹Ö‚PÁJÕ—Mlá?Ù h,œ6b«éʸ6–¬RÖª5 ”NØ› Ûµ±xo¡^…6ÁË2púT4Ì1S~y±fµC€óJÓ ¬D7•É÷° Àâ¾èø–p«{ '‹†1Lø!ÌiçÙÅ“:—^+¾¥7åˆÃò‡-]f>AŽµ{A¼}¼geGNÇ7O4<ÊvzÊæ5Ê^D¦l)šãRâb©|^ÊHTn,¿ ^ äó¹Ó4IÞ—M3/ק{°žE_k¹¥`Tý.h맱å=”¯Ði JB¥Ì ʇR*s@#䎥ðmlÙÇrÝMІQ½7TX•NÂL„²[LµwЀG’ìDW:EâÎzàÓŠRÅâ <š…wû¯J¯²ÓÒx,î _V¢hf²¯<Ÿx¬)ÏwÛÁ¯áúÇ2|n£¦LH‹Þ1‰©wT®}GÌÄ‚`@̤I̼´õiÄÌI¦b™È$ù¹´Wƒ¡¼Fn³-ó<–óæ\²[èJŸ1å‹8*Š:Ž¤‹'ç3$¶Ÿ²ÆØñ¹ˆåø$”aEž{sÄ ÿSýš †R|K“[Ì%jHq*sQ¦d<ÿSüt‹Òó”{ëÎÝY–›SÅ å¸Z!®–ˆ«ƒã««æ’ü8.¹{ò¸æv“KêçßÑ‚Bn±ŠÃæ¦ÜUÌBm[U$xKÓhúÄëX1,¥y<¤)>3ùgª¿+3*X,o¸–™ˆ”4%¤Ï÷¢ ”iKÈѬªû²êªUG;7 ý×ÿjèy²‡OŒîô‰ÉÉGv(ɱ!ÔÆÅ«¨ä–9ô²4‡ÈIã§o?Ø‹í_me…8ƒáÍþn){´SÊ^‡‰ä? FÞ!¿å¨£7K“ã×ÂV4;Ž­Š¦f‰Jš‚C ›>"t'á}w7ü4W‰&IÑ´éO%âÇ´òkp•#¸ ÖfWª®ŽuÑÞAäÙ9S,\$O#‡ªN~+úLÊ‘öÓ[éVº‡ÒÄÈsë¿2ÓtÙ‚ûT³³8©ºwl‡¥Y$’U3G %(?¢¤ §+ˆ£Ú°¾ °ôTÎDz%óÕ¨ÎÝvf ~ ®îšw0l‡í`Šlqš-Þ€&ZívšU#¥§ù¨£Ú"ózÔe>àܬº3ŒCѽãÎCXœá#‹ôWpý*®S7‡\ #ÒiS¥’-“a(xýª]ñʧöeñHg˜Í \¢Ì»¡‹Qd­ö«~‡ºYnÉÚM¬è_38GÁ¹š!hšlíòˆ¶³ƒÆaRÑ{«•Gì•p‰î½½A¨ íyÆ ä©¶›¸ëff¼¹áì›™Ã,£lfnÙ=AUœ´Åµ‡¢ÇJ[Üm"çìôt[Ce˜¨„DÙŒ) PJX”鶈äºB¬<¬ÈÛ:–Øo–ä$d;†Doæ3œáȶ¢=ÜüÝeei)b[)§qO?x3Ž\æ‹è8ˆÒ;J¹>{”’ùí–Îu~åkd;Î:sT¶Ÿ£ôDaW‚Ò° 1>VàgséXÞ9Ë![,«ø­=ßqè<(ˆÀnVlßþ¤CúYó1Œ´ÈÆcVmP*/1AET¤w4bVi‡Nõq÷ªK8˜.¸B÷rHˆöîÖ/(bG0>¤žU6‘ʘHHNDKðaòäjB¹Jäû,’8‚•“,bdl‰,‚Úþs·Ó,9–HØÁ€“Ê‘ÌP`Ìðñm®¡°¹&O ¾)H5×$äô»û¦|×Í+@"¬åa0Ì3yºõLã2i>ÇÊ4n®kxÕNp˜;‡Ç'í›–;9r{=tkNƒàÚ: â„,ˆ÷Ðœ“ƒ›Xn ¹Ûˆ»GͤãÔgvŒÎkpåÈØ^°¹?çIaø¾.)H_8™=RÆÞ"( L8ëÿõ>å‡ìT~`±sbÂÇžÛÓÈ@ òÎ{“­oÕÎÞ°\Îí[fLÏ¡|(Ké¯Aj ‡Ìêé!%†µìŒý¦ó©7­0@tyÈçºeÞ•¥D΃ Èî[ÌÄ‚ ïÁÆppÏ(î)nóXâì5ŒîÇrë@õ"–‡ç·(>7‰ÛÁ¼v<’ÓÐ"t7ÁSÉ£X€·D¯ 3|DdºŽyD‡Ëo×¥‡ äÕô5r;éÆIˆG¹HÝíÄzGïÜ=²Ñû=”/|y‚Ñëã`ø“ Ïfþ¾‹{¼ówÌKeGºÊ¤÷lV)Dz‘{JX^5œa'Aïíhgâ2PÇÌÝ>·¶¤žpªÂExÆÈnsÈ(?ÇÊÔ¿W¸Üá — WøIÒ(Nïb.ê}_ÛˆÛÔR},Þ¹a3«i9ŸHùJ¾OS;]SÝ(HåübˆDf ٸܟ°;Ž¯#3X…ÎÖC^W[¾¯åÊE’.‘Ï^¢£ÆVH~ÌÐ!˛Ұ re‡™õq-]¹ySêì¤_µáa×aŽe`¨” ÆöûÞ™çh‹sIÎ]yZ7N¼äYoPääbxšFæiíÝ÷Hž¢ÝedècíRÞÀŽÊåÛ'ç©Ò)[?9OÆÊ‹Í¢ÚpTj ÞåôxWq¯¥?¹HNK— ¾;(‡¸@®ýñ­Ÿ pu:³Þ˜ÍùÅð7û?ÿEøû7w|OŠa¼O*s$oë$õÙ"ã˜1°é(N H‡…ÈT×£éIF»š¦2¸È+2ó.½(R_t7¿>  ˆu°¥ž‹ÝD…*¿d?*”É7í¨@¼Ÿþðà3`C(D ”>ìÓéå«ÊüùÒŠNº¾­Ô¶ï†óþèÕ´8ùA3:+ƒ¿¿†‡ß&vð„õd¸Ý#ûžO(¶>¡Øì[ŸPÃ}ŸÐXaO!XSÆNú/VÖˆqÕ_•¿óœG÷oíJ¾C{Ib‚|lSÝå1ëšA[4Nâ®ÿA‘{d ¯caÍÇÎo$íü÷FVghssä_;9A…‰‰ê’sÚ3×O{>š˜ÿs%'Cò¡d/$TŠÚ‰m>v¥¶J„ɬ«@G–Ú¾ÃR3ÍB1—·µ˜‹ëàv‹\kùÃm询Q¢Rnÿ¶™´¿­ùr?¾ùê;*Ad endstream endobj 1087 0 obj << /Type /Page /Contents 1088 0 R /Resources 1086 0 R /MediaBox [0 0 612 792] /Parent 954 0 R /Annots [ 1083 0 R 1084 0 R 1085 0 R ] >> endobj 1083 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [39.433 681.785 51.396 690.225] /A << /S /GoTo /D (noweb.NW2i2LRH-3dLroe-1) >> >> endobj 1084 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.862 679.148 166.825 692.213] /A << /S /GoTo /D (noweb.NW2i2LRH-3dLroe-1) >> >> endobj 1085 0 obj << /Type /Annot /Subtype /Link /Border[0 0 1]/H/I/C[1 0 0] /Rect [154.478 619.608 165.939 631.713] /A << /S /GoTo /D (noweb.NW2i2LRH-3Gkbo6-1) >> >> endobj 1089 0 obj << /D [1087 0 R /XYZ 71 791.866 null] >> endobj 362 0 obj << /D [1087 0 R /XYZ 81.963 682.871 null] >> endobj 1090 0 obj << /D [1087 0 R /XYZ 72 680.144 null] >> endobj 1091 0 obj << /D [1087 0 R /XYZ 83.504 682.871 null] >> endobj 1092 0 obj << /D [1087 0 R /XYZ 83.504 670.916 null] >> endobj 1093 0 obj << /D [1087 0 R /XYZ 83.504 658.961 null] >> endobj 1094 0 obj << /D [1087 0 R /XYZ 83.504 647.006 null] >> endobj 1095 0 obj << /D [1087 0 R /XYZ 83.504 635.05 null] >> endobj 1096 0 obj << /D [1087 0 R /XYZ 83.504 623.095 null] >> endobj 1097 0 obj << /D [1087 0 R /XYZ 83.504 611.14 null] >> endobj 1098 0 obj << /D [1087 0 R /XYZ 83.504 599.185 null] >> endobj 1099 0 obj << /D [1087 0 R /XYZ 83.504 587.23 null] >> endobj 1100 0 obj << /D [1087 0 R /XYZ 83.504 575.275 null] >> endobj 1101 0 obj << /D [1087 0 R /XYZ 83.504 563.319 null] >> endobj 1102 0 obj << /D [1087 0 R /XYZ 83.504 551.364 null] >> endobj 1103 0 obj << /D [1087 0 R /XYZ 83.504 539.409 null] >> endobj 1104 0 obj << /D [1087 0 R /XYZ 83.504 527.454 null] >> endobj 1105 0 obj << /D [1087 0 R /XYZ 83.504 515.499 null] >> endobj 1106 0 obj << /D [1087 0 R /XYZ 83.504 503.544 null] >> endobj 1107 0 obj << /D [1087 0 R /XYZ 83.504 491.588 null] >> endobj 1108 0 obj << /D [1087 0 R /XYZ 83.504 479.633 null] >> endobj 1109 0 obj << /D [1087 0 R /XYZ 83.504 467.678 null] >> endobj 1110 0 obj << /D [1087 0 R /XYZ 83.504 455.723 null] >> endobj 1111 0 obj << /D [1087 0 R /XYZ 83.504 443.768 null] >> endobj 1112 0 obj << /D [1087 0 R /XYZ 83.504 431.813 null] >> endobj 1113 0 obj << /D [1087 0 R /XYZ 83.504 419.857 null] >> endobj 1114 0 obj << /D [1087 0 R /XYZ 83.504 407.902 null] >> endobj 1115 0 obj << /D [1087 0 R /XYZ 83.504 395.947 null] >> endobj 1116 0 obj << /D [1087 0 R /XYZ 83.504 383.992 null] >> endobj 1117 0 obj << /D [1087 0 R /XYZ 83.504 372.037 null] >> endobj 1118 0 obj << /D [1087 0 R /XYZ 83.504 360.082 null] >> endobj 1119 0 obj << /D [1087 0 R /XYZ 83.504 348.126 null] >> endobj 1120 0 obj << /D [1087 0 R /XYZ 83.504 336.171 null] >> endobj 1121 0 obj << /D [1087 0 R /XYZ 83.504 324.216 null] >> endobj 1122 0 obj << /D [1087 0 R /XYZ 83.504 312.261 null] >> endobj 1123 0 obj << /D [1087 0 R /XYZ 83.504 300.306 null] >> endobj 1124 0 obj << /D [1087 0 R /XYZ 83.504 288.351 null] >> endobj 1125 0 obj << /D [1087 0 R /XYZ 83.504 276.395 null] >> endobj 1126 0 obj << /D [1087 0 R /XYZ 83.504 264.44 null] >> endobj 1127 0 obj << /D [1087 0 R /XYZ 83.504 252.485 null] >> endobj 1128 0 obj << /D [1087 0 R /XYZ 83.504 240.53 null] >> endobj 1129 0 obj << /D [1087 0 R /XYZ 83.504 228.575 null] >> endobj 1130 0 obj << /D [1087 0 R /XYZ 83.504 216.62 null] >> endobj 1131 0 obj << /D [1087 0 R /XYZ 83.504 204.664 null] >> endobj 1132 0 obj << /D [1087 0 R /XYZ 83.504 192.709 null] >> endobj 1133 0 obj << /D [1087 0 R /XYZ 83.504 180.754 null] >> endobj 1134 0 obj << /D [1087 0 R /XYZ 83.504 168.799 null] >> endobj 1135 0 obj << /D [1087 0 R /XYZ 83.504 156.844 null] >> endobj 1136 0 obj << /D [1087 0 R /XYZ 83.504 144.889 null] >> endobj 1137 0 obj << /D [1087 0 R /XYZ 83.504 132.933 null] >> endobj 1138 0 obj << /D [1087 0 R /XYZ 83.504 120.978 null] >> endobj 1139 0 obj << /D [1087 0 R /XYZ 83.504 109.023 null] >> endobj 1086 0 obj << /Font << /F18 190 0 R /F57 146 0 R /F73 223 0 R /F61 150 0 R /F55 149 0 R >> /ProcSet [ /PDF /Text ] >> endobj 1140 0 obj << /Length 147 /Filter /FlateDecode >> stream xÚ%‰± Â0E_éö–|‚÷ L“h¡S V0ƒ “ƒ8YGAEçöÓâŸäÜ2ˆ1(œáÜsu=Õ¨` æЦF¯øʪɱ‚VÿçxæÖ±ÜA5,W9³tkÜoËv³@Þö Õ]É~hH”Ò'>–!áRÊ”O“)Qù&ñ ö†Ò,¼t¼å/é÷3z endstream endobj 1082 0 obj << /Type /Font /Subtype /Type3 /Name /F148 /FontMatrix [0.02007 0 0 0.02007 0 0] /FontBBox [ 3 15 23 36 ] /Resources << /ProcSet [ /PDF /ImageB ] >> /FirstChar 178 /LastChar 178 /Widths 1141 0 R /Encoding 1142 0 R /CharProcs 1143 0 R >> endobj 1141 0 obj [26.46 ] endobj 1142 0 obj << /Type /Encoding /Differences [178/a178] >> endobj 1143 0 obj << /a178 1140 0 R >> endobj 1144 0 obj << /Length 187 /Filter /FlateDecode >> stream xÚmÍ!Â@…á%+šŒ©ÅuN@[(MQM $T€B °Ð£õ({„•ÒÇLóMöb³|4+8á) ÎÆœg|JéJ“\bÂrúåx¡yMñNÅ«ÞzÍ÷ÛãLñ|³à”âŠ÷)'ª+6èŒ g"àmJÈû §AcÑÂ˜Ò ¡xÁú§2TÌ/:!RÊV@##´wš>š¼&#í¯ÏÀ à­ÐZ/ÉzýÜã:8ZÖ´¥/•;hr endstream endobj 1081 0 obj << /Type /Font /Subtype /Type3 /Name /F144 /FontMatrix [0.011 0 0 0.011 0 0] /FontBBox [ 5 28 42 64 ] /Resources << /ProcSet [ /PDF /ImageB ] >> /FirstChar 178 /LastChar 178 /Widths 1145 0 R /Encoding 1146 0 R /CharProcs 1147 0 R >> endobj 1145 0 obj [46.98 ] endobj 1146 0 obj << /Type /Encoding /Differences [178/a178] >> endobj 1147 0 obj << /a178 1144 0 R >> endobj 1148 0 obj [999.5 748.3 794.7 744 1060.5 547.8 632.9 990.9 869.4 1171 935.7 860.6 807.5 809.6 877.7 673.5 724.2 867.9 812.1 1178.9 871 734.3 803.2 0 0 0 936.9 936.9 0 0 0 0 0 0 422.7 422.7 0 0 302.1] endobj 1149 0 obj [847.7 652.5 845 712.6 794.4 973.6 781.7 959.9 714.7 700.1 646 584.1 587.7 523.7 651.6 648.9 354.1 583.2 596.3 646.3 638.5 588 664.4 610.9 700.1 580.8 621.9 873.6 587.7 821.8] endobj 1150 0 obj [990.3 777.2 987.4 842.8 932.2 1127.7 918.2 1112.7 845.1 829.2 770.1 702.5 706.4 636.6 776.3 773.3 451.5 701.6 715.8 770.4 762 706.8 790.2 731.8 829.2 699 743.8 1018.6 706.4 962] endobj 1151 0 obj [666.7 611.1 777.8 777.8 388.9 500 777.8 666.7 944.4 722.2 777.8 611.1 777.8 722.2 555.6 666.7 722.2 722.2] endobj 1152 0 obj [777.8 777.8 777.8 777.8 777.8 777.8 777.8 777.8 944.4 500 722.2 777.8 777.8 500 500 722.2 722.2 722.2 777.8 777.8 777.8 777.8 777.8 750 1000 1000 833.3] endobj 1153 0 obj [805.6 805.6 0 0 0 0 0 0 0 0 1000 1000 1000 1000 0 0 0 0 1000 1000 500 500 1000 1000 1000 0 1000 1000 666.7 666.7 1000 1000 1000 0 0 1000 0 0 0 0 0 0 888.9 888.9 888.9 888.9 888.9 888.9 888.9 0 0 0 0 0 0 0 0 0 527.8 444.4 0 0 0 0 0 0 1055.6 1055.6 527.8 0 0 0 0 0 1444.4 1411.1 444.4 0 0 0 0 0 1055.6 1411.1 0 0 0 0 0 0 0 0 0 0 500 666.7 0 0 0 0 0 0 0 0 0 0 0 0 450 450 450 450] endobj 1154 0 obj [585.3 892.9 585.3 892.9 892.9 892.9 892.9 892.9 892.9 892.9 1138.9 585.3 585.3 892.9 892.9 892.9 892.9 892.9 892.9 892.9 892.9 892.9 892.9 892.9 892.9 1138.9 1138.9 892.9 892.9 1138.9 1138.9 585.3 585.3 1138.9 1138.9 1138.9 892.9 1138.9 1138.9 708.3 708.3 1138.9 1138.9 1138.9 892.9 329.4 1138.9 769.8 769.8 1015.9 1015.9 0 0 646.8 646.8 769.8 585.3 831.4 831.4 892.9 892.9 708.3 917.6 753.4 620.2 889.5 616.1 818.4 688.5 978.6 646.5 782.1 871.7 791.7 1342.7 935.6 905.8 809.2 935.9 981 702.2 647.8 717.8 719.9 1135.1 818.9 764.4 823.1 769.8 769.8 769.8 769.8 769.8 708.3 708.3 523.8 523.8 523.8 523.8 585.3 585.3 462.3 462.3 339.3 585.3 585.3 708.3 585.3 339.3 938.5] endobj 1155 0 obj [277.8 777.8 500 777.8 500 777.8 777.8 777.8 777.8 777.8 777.8 777.8 1000 500 500 777.8 777.8 777.8 777.8 777.8 777.8 777.8 777.8 777.8 777.8 777.8 777.8 1000 1000 777.8 777.8 1000 1000 500 500 1000 1000 1000 777.8 1000 1000 611.1 611.1 1000 1000 1000 777.8 275 1000 666.7 666.7 888.9 888.9 0 0 555.6 555.6 666.7 500 722.2 722.2 777.8 777.8 611.1 798.5 656.8 526.5 771.4 527.8 718.7 594.9 844.5 544.5 677.8 762 689.7 1200.9 820.5 796.1 695.6 816.7 847.5 605.6 544.6 625.8 612.8 987.8 713.3 668.3 724.7 666.7 666.7 666.7 666.7 666.7 611.1 611.1 444.4 444.4 444.4 444.4 500 500 388.9 388.9 277.8 500 500 611.1 500 277.8 833.3] endobj 1156 0 obj << /Length 323 /Filter /FlateDecode >> stream xÚ¿JÄ@Æ¿ba›LR‚|z5/jóàO¤†Ôž†õAAMQ¯­•ÝØ‚-+^'èiHv!æc"NVÊÒ”äC¼ŠA}Þéký <ëÎ endstream endobj 1157 0 obj << /Length 259 /Filter /FlateDecode >> stream xÚÅѽNÃ@ pW"yÉ#ä^.)j#¦“J‘È€êŒ ˜G»GÉ#dÌPÅü}ú±°0p‰ô;%ŽcûVåe]¹Ò]¹‹Eí–+·\¸×Š?–¸êëéÕË;¯öO`‡çì›{÷õùýÆ~ýpã*ö÷ŒT[n6ŽˆŠžHFÒ‘ê\ÕÈU#ªÕ` „}„ŠR>‚eÆ@3|(Het %‰Ø¶vSˆ§ ÈÏéOÉþÆy2ûC{ ˜ˆ©Îp¤ÛW=õÐïh®ÇnG*Ò"bIˆ ‹$ØèH0Ad—LO2³ºDZÿŠØ9ÛI[•¿2EòmÃü¬^µô endstream endobj 221 0 obj << /Type /Font /Subtype /Type3 /Name /F87 /FontMatrix [0.00836 0 0 0.00836 0 0] /FontBBox [ 3 -29 59 52 ] /Resources << /ProcSet [ /PDF /ImageB ] >> /FirstChar 103 /LastChar 112 /Widths 1158 0 R /Encoding 1159 0 R /CharProcs 1160 0 R >> endobj 1158 0 obj [60.71 0 0 0 0 0 0 0 0 60.71 ] endobj 1159 0 obj << /Type /Encoding /Differences [103/a103 104/.notdef 112/a112] >> endobj 1160 0 obj << /a103 1156 0 R /a112 1157 0 R >> endobj 1161 0 obj [478.7 478.7 358.7 876.3 358.7 876.3 358.7 601.1 601.1 601.1 601.1 601.1 601.1 601.1 601.1 601.1 601.1 601.1 292.1 292.1 0 876.3] endobj 1162 0 obj [458.3 458.3 416.7 416.7 472.2 472.2 472.2 472.2 583.3 583.3 472.2 472.2 333.3 555.6 577.8 577.8 597.2 597.2 736.1 736.1 527.8 527.8 583.3 583.3 583.3 583.3 750 750 750 750 1044.4 1044.4 791.7 791.7 583.3 583.3 638.9 638.9 638.9 638.9 805.6 805.6 805.6 805.6 1277.8 1277.8 811.1 811.1 875 875 666.7 666.7 666.7 666.7 666.7 666.7 888.9 888.9 888.9 888.9 888.9 888.9 888.9 666.7 875 875 875 875 611.1 611.1 833.3 1111.1 472.2 555.6 1111.1 1511.1 1111.1 1511.1 1111.1 1511.1 1055.6 944.4 472.2 833.3 833.3 833.3 833.3 833.3 1444.4 1277.8 555.6 1111.1 1111.1 1111.1 1111.1 1111.1 944.4 1277.8 555.6 1000 1444.4 555.6 1000 1444.4 472.2 472.2 527.8 527.8 527.8 527.8 666.7 666.7 1000 1000] endobj 1163 0 obj [830.5 608.2 649.3 604.4 884.5 430.8 506.1 822.9 715.4 982.3 774.1 707.6 660.6 662.5 722.7 542.1 586.9 714.1 664.7 989.3 716.8 595.8 656.8 0 0 0 775.1 775.1 0 0 0 0 0 0 320 320 0 0 213.4] endobj 1164 0 obj [388.6 388.6 277.6 756.2 277.6 756.2 277.6 501.8 501.8 501.8 501.8 501.8 501.8 501.8 501.8 501.8 501.8 501.8 216.1 216.1 0 756.2 0 362.4 0 717.6 884 612.5 831.6 662.8 611.2 785.4 720.3 554.5 552.3 668.4 666 1049.5 832.1 827 827.5 827 828.1 828.6 669 645.8 831.3 1046 719.2 832.9 602 277.6 0 277.6] endobj 1165 0 obj [545.3 867 917.1 932 734.6 877.7 791.6 870.4 1002.5 855.7 1050.4 805.1 809.6 748.6 627.1 682.7 639.6 694 687.3 439.5 688.5 673.2 759 738.2 686 749.2 680.2 744.9 641.8 724.5 883.2 712.6 913.4 1022.6 610.7 684.8 1050.7 0 0 923.5 0 0 0 0 0 0 0 0 625.6 625.6 625.6 625.6 625.6 625.6 625.6 625.6 625.6 625.6 374.8 374.8 915.5 628.1 915.5 0 694.3 931.7 801.7 868.5 997 744.6 625.3 925.9 946 506.6 515.4 816.7 692.7 1240.8 998.5 968.6 712 997.6 749.2 690.6 617.9 936 791.2 1175.7 813.6 688.3 814.2 0 0 0 0 0 499.9 749.2 726 611.3 743.4 625.6 535.6 703.7 763.5 467.8 436.1 689.1 473.9 1095.9 811.8 697.6 727.2 745.2 549.3 576.2 531.9 787 621 978.7 655.5 732.1 593] endobj 1166 0 obj [428.6 713.3 757.6 770.8 596.1 722.7 646.6 716.3 833.2 703.3 875.6 658.5 662.5 608.5 501 550.2 512.1 560.1 554.2 334.9 555.3 541.8 617.7 599.3 553.1 609 548 605.3 514 587.2 727.6 576.6 754.3 851 486.4 552 875.9 0 0 763.2 0 0 0 0 0 0 0 0 499.6 499.6 499.6 499.6 499.6 499.6 499.6 499.6 499.6 499.6 277.6 277.6 756.2 501.8 756.2 0 560.4 770.5 655.5 714.6 828.3 605 499.4 765.4 783.2 394.3 402.1 668.7 559.1 1044.1 829.7 803.2 576.1 828.9 609 557.2 492.9 774.3 646.2 986.5 666 555.1 666.6 0 0 0 0 0 388.4 609 588.5 487 603.9 499.6 420 568.8 621.7 360 331.9 555.8 365.4 915.8 664.4 563.4 589.6 605.5 432.1 455.9 416.7 642.5 495.6 812.1 526.1 593.9 470.7] endobj 1167 0 obj << /Length 197 /Filter /FlateDecode >> stream xÚŽ± ‚`…4wéº/Pj–)‚äÔÔMÕØPÔš>šâ#46È_Gth ¾åÞË=監(TW½©Cõ# |=yr•±Ï­«3¯;/’fâìt싳â^œl­÷Ûã,Nº™+ç…î=u’-ˆ',ƒž]£ÿÆ BR"/𠬽wƒý‚]¡OJ HÑ4äMJ‡êÿ0?_9º¨¤èÂÙÂ.6²—í­†Õ¾Ñ†ô¤-iN«Í‹e™ÉV¾‘L endstream endobj 1168 0 obj << /Length 196 /Filter /FlateDecode >> stream xÚ= Â@…Ÿ¤¦Éœ èæGb¬à‚Vb¥–Š‚•ñh%GH™"¨/ÙÂVøšeÞûf˜ Æ©šj?Õ8Ò$ÖC(g‰b…îg’©³Ñ(³àXŒ]êõr;Š™®fÊw¦ÛPƒØLï@ |Èù “~‰n¯Fç ˜<z/ø¤@—”ð:ŽºMrpíñïß\3]8[ØÅFö²ÝiäÍÝhHOÚÒ™æôÿ´A¼HæVÖòxuO’ endstream endobj 1169 0 obj << /Length 197 /Filter /FlateDecode >> stream xÚ=;Â0DQ m“#°'À1ùQ. ¢@T@I‚Ž–£pʈ°'XÊÈzã™ì&Ùp:áˆSN8qjøhèBq&,âd ãp¦Â’Þrœ‘^ %mW|»ÞO¤‹õœ é’w†£=Ù’•œ\¾à%Ò‹„NfN‚¦Rª×8þÔŽ;Óó§„À?]¨AÈq„Àë¶ !帿ÁÅ;$E‹ïC3þÑóNÕM€YBï¨vÒ¶ò¿6Ân*§…¥ ýUKe endstream endobj 1170 0 obj << /Length 97 /Filter /FlateDecode >> stream xÚ31Ó³´P0P02PеP0²P03VH1ä*ä2 (˜Be’s¹œ<¹ôÃ̹ô=€Â\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹BMÍ?ÊAM —«'W (»V‹ endstream endobj 1171 0 obj << /Length 194 /Filter /FlateDecode >> stream xÚE1Â0AH×ä Ü °MœR. ¢@T@I‰Ž<§ä ))æ¼ àbeÝx½{6íG¬9aËvÀ‰á½¡Å©Ì4Û!ÀîH™#µæ8%5—))·àËùz •-§lHå¼1¬·ärÖ-9· ï ò¹—"“§HôéÒöEÀ •H$;5ÆšÀøÿ2à¨úÞ ïà€¿Ôó¢É@L¾lº ú¡)åa7lI3G+úùlJ endstream endobj 1172 0 obj << /Length 110 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0´TеP01Q03VH1ä*ä22Š(˜B¥’s¹œ<¹ôÃŒL¸ô=€â\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹BýÿÿQÀ¿? C ýGõÐG\®ž\\0ñoy endstream endobj 1173 0 obj << /Length 112 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0VеP0²P03VH1ä*ä22 (˜Bd’s¹œ<¹ôÃŒL¸ô=€Â\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹Býÿÿ‘@ýÿÿ öC Õÿÿê…\®ž\\¼HB€ endstream endobj 1174 0 obj << /Length 180 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0UеP01R03VH1ä*ä26 (˜Bd’s¹œ<¹ôÌ͸ô=€Â\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹ƒýæÿøA„<ˆ¨‡@ føÄŒ@ÄÚÌ ‚ýÈa@Âìˆÿ@D9‰eÁê :˜ÐÉ‘àЇ $„ÀaU?pH‚ú\®ž\\Äoˆ endstream endobj 1175 0 obj << /Length 179 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0UеP01R03VH1ä*ä26 (˜Bd’s¹œ<¹ôÌ͸ô=€Â\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹Bý†ÿ@Ä0Ñ"00Ôƒ†ú@‚D0ÿŒô @Žaø"þƒW"íÿÝ "Øá,°X¬îL/=  p¸Ô€†ÕÑ ISÀ¡ËåêÉÈŽõp endstream endobj 1176 0 obj << /Length 143 /Filter /FlateDecode >> stream xÚ31Ó³´P0P04S02U06V05SH1ä*ä ¡±!T*9—ËÉ“K?ÈåÒ÷Šsé{ú*”•¦ré;8+ù. ц ±\ž. ìø?Èÿ°ÿ„ÿ€ð¿üþÿìÿì¡°ž¡ÿ1üaüÃüƒùPíûÿüoøÏPÃ`ÁÀåêÉȪL.D endstream endobj 1177 0 obj << /Length 142 /Filter /FlateDecode >> stream xÚ31Ó³´P0P04SÐ54V06R04TH1ä*ä24 (™Àä’s¹œ<¹ôà M¹ô=€\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. ìø?Èÿ°ÿaÿÿÙÿ“ÿÇÿýCÃ?†?Œ@Èüƒÿƒý‡ÿþøßPÇ`ÁÀåêÉÈÑ4,r endstream endobj 1178 0 obj << /Length 96 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0Æ@ÚP!Å« H(€¹`™ä\.'O.ýp —¾˜ôôU()*MåÒw pVò]¢zb¹<]äìêüƒõìä¸\=¹¹ŠŽ– endstream endobj 1179 0 obj << /Length 162 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0UеP01R03VH1ä*ä26 (˜Bd’s¹œ<¹ôÌ͸ô=€Â\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹<bÆ@‚N°ƒ ?˜8$äÁD° ­õ ¢Dü`#˜ø2î©fâ2˜X3Iq,63© ã*@—«'W yK/ endstream endobj 1180 0 obj << /Length 104 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0Æ@d©bÈUÈeh䃹`™ä\.'O.ýpCC.} 0—¾§¯BIQi*—¾S€³PÔE!¨'–ËÓEAžÁ¾¡þÀÿ0XÀ¾Až0À®ËÕ“+ 9Ü-I endstream endobj 1181 0 obj << /Length 111 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0V04W01Q0¶PH1ä*ä21PA#CˆLr.—“'—~¸‚‰—¾P˜KßÓW¡¤¨4•Kß)ÀYÈwQˆ6T0ˆåòtQ°ÿÿÿÿŸz ñï?*‹1pš¶ƒËÕ“+ ÏJS endstream endobj 1182 0 obj << /Length 254 /Filter /FlateDecode >> stream xÚ1NÄ0E¿•"Ò49BæDÉ.¡!Ò²H¤@‚ŠQ±” èV»9šâ#X¢qabfô4ÏÒ·ìùoºõéyÏ5w|Òp×òêŒw ½R«aÍ«~¹yz¡ÍHÕ=·=U×S5ÞðûÛÇ3U›ÛKn¨ÚòCÃõ#[J  K(R’(¥ ƒâè‹4ÎÌÈ]¹7Ÿ^Ž/_Ø2 †Êé(€`À\„ Ä<LÈâÁx÷ÙÄ\Xìÿ…¿gúAø…7A¿÷:ÈëH§ÃÖ°ZÈjµIKjë)î! "c¢hI(‚QUƒJ{Õ·²1Gše/¥]tGß¹<› endstream endobj 1183 0 obj << /Length 190 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bSSs…C®B.ßÄ1’s¹œ<¹ôÃLŒ¸ô=€¢\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. ö P߀ ÿÀJ2~~€‡dþü|"ÙþÀN‘üþ`%åê°’ö õ ìhL²¨ FÖÿÿ'ÿÿy“ü´ñû,$3üÀÀŒêÿ3Øÿo€ÿAŒYœËÕ“+ H0‚6 endstream endobj 1184 0 obj << /Length 198 /Filter /FlateDecode >> stream xÚ}ϱ Â0à” …[|„Þ˜TkÑI¨Ì èä Nêè èj}´>J¡c†Ò˜ 4è „|4ùÓ;.ˇ³)J¹q’ã)…+di¹#Ç  Ä3 bånA¨5Þo3ˆb³ÀD‰ûåT‰•µšYk[Âz^Dí«yÃ’Æ1 ¸é‰<­ƒý§QøƒS˜H¨hUsjD0NÍû/ëýÕò£QG<ó¿Tá]×ãKÅDbh@C£6„“3K[x£Gœj endstream endobj 1185 0 obj << /Length 230 /Filter /FlateDecode >> stream xڥбJÄ@Æñ/¤L³°óº Éi¬ç ¦´²+µ´P®Û©¸©ysÎϽQ­‡%oÚõæ镶=¹{®[r×é˜\ÃïŸ/䶷—\‘ÛñCÅå#õ;üÐ"ÓL EÐÅ(ðJ£däG)‚2£3!_±#2±C¢[°â•Ã{GEþÁòÀá{ÿûåÊ° :Z2 fFŠ…€bÖ˜9eÙ)úQSFÊO?˜žV2C—ºê鎾?9ru endstream endobj 1186 0 obj << /Length 197 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bS3…C®B.C ßÄI$çr9yré‡+˜ré{E¹ô=}JŠJS¹ôœ€¢. Ñ@-±\ž. õÿÿ700üÿÿÿ˜ü&ÿÉÿ @Y 4ûÆÿ€$ƒý)&ù?€Hö ’L2þA ÿÈ:0Y&íq‘ Ržbb¦ùõH.©C¸ ÙÍ_@|ñü¸¯A! HÈÀCé,ô !ÉåêÉÈ݈I endstream endobj 1187 0 obj << /Length 149 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bSS3…C®B.C ßÄI$çr9yré‡+˜ré{E¹ô=}JŠJS¹ôœ€¢. Ñ@-±\ž. õÿÿÿÿÃ$þÿÃBÖƒIæ uD“6`’ùD2þÀJþÿO˜Ä¥j2ÛøÁ¤|©$(4þ7üÇA‚e¸\=¹¹WD–Ü endstream endobj 1188 0 obj << /Length 230 /Filter /FlateDecode >> stream xÚ•½NÃ0€/Êé?Bî¨Õ…vŠÔ©`b@LбÖÞ Þ$R_à¤.•åjŸKÅŠ-}ƒo¸ï³»ͦTИ®JrŽ&7ôRâ+º‚ÂLãäy‹óí¹íÚ?£­oéýícƒv~· í’K*ž°^ÀdòÃ`dÏPÉ‘¡aD¾„ÓZN{¨8;@Ά:0œGdœzT €”Û š3’Î#ìò§ÿ'dè!Q M„4·éò³† ýú™¨«â>¡Š1š¥š£5ßȎ×t*Œ«ïñïć5 endstream endobj 1189 0 obj << /Length 105 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bS3…C®B.c# ÌI$çr9yré‡+qé{E¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]êÿÿÿ÷ €ÄØ0 %ŒË\®ž\\6Qg? endstream endobj 1190 0 obj << /Length 126 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bS3…C®B.rAɹ\Nž\úá &\ú@Q.}O_…’¢ÒT.}§gC.}…hCƒX.O…úÿêÿÿ``ø'ê!P‚È:„ˆ°'–¨ÿ`àbÿ¸\=¹¹”…jo endstream endobj 1191 0 obj << /Length 201 /Filter /FlateDecode >> stream xÚ­Ð1 Â@ЋÀ4¹Î \kP1‚)­,ÄJ--!9™D,,½î¶T×]…S[̃ù3Õçn¥Q§*9z¸K5—¦.s½WÍj“9ú!²!qŽ¬«SdaVËõ ™ßo“ƒ, ‘þcP$”ˆnPPB¥z@QÉÈ(>Z°öðãlíl/5.§žâÒ×ÄK=&M£Ø¥ÄÆ(o9)÷Œ[‘·•ä-Ç_m0ÂÍv¢ž`é®Þfsì„8À„‰‰ endstream endobj 1192 0 obj << /Length 199 /Filter /FlateDecode >> stream xÚeÈ1 Â0Æñá-ÁwÓ–ZtP¡*ØAÐÉAœÔÑAQPPRo¦7iqpT· ¥±I( 8¼ïû{~£ÝBÝâ¼&6}\9°Ol[Lñ,7„@gè¹@GEŽq¿;¬“>:@8wÐ^@8@–’X&äaüÆs!Ëe—V^Äz“ÒHø4½ ¦±6Q¾µ±25> stream xÚÍ=‚@…ÇXLØ è²ò#V› &na¢•…±RK ¶ÂÑö(Ò‚°.CÇ l¾â½æ}/Nç«ŒBJh&)^P²¤«ÄF© ]œ Í厹Fq¤(E±u1 ½£×ó}C‘ï×$Qt’žQʘZÛ‚gm¾µ‡J9Õ€ƒeÊ ‚ºðë€7FçдÓ?oaŒFòú½ k©ïÄ öª,Íú|_ÂçàFã&h endstream endobj 1194 0 obj << /Length 182 /Filter /FlateDecode >> stream xڭϱ Â0àHá^Â{ÓZìZ+˜AÐÉAœÔÑAÑÕöÑú(}„Ž$ç] 8Nù ¹ä¿KšMó9&èx¥gž,\Áå¼Od+…ã f‡.³âS0~÷Ûã ¦Ø,Ђ)qo19€/±"jõB¨.P«;UuÌD÷ŒF¯ Âó'a¸üy£bhŒçF±¥4j-iMËðO*ªÿ·ù"`éa oõÆ…t endstream endobj 1195 0 obj << /Length 217 /Filter /FlateDecode >> stream xÚÅÐ1nÂ@Ð(,Mã#ì\€¬ƒj%C¤¸ˆ* ”*I™"Z¯o+íMð|k7c£Ü€æšùž?¬Wœð‚giÆóG^,ù#¥oÊrI^Ž§÷/*JÒ{ÎrÒÏ’“._øt<’.^7œ’Þò!åäÊ-Çi¢<â”PÁ™‚ƒq¶†j,ÿQÑ?¼ÐvÀôÞtÂDððQ[ òŸ½ba~®ÔêÒ§ú‚±ÐWJõ‡!†Id“øö†©G”‰›‘ÈÑSI;úŒ%  endstream endobj 1196 0 obj << /Length 218 /Filter /FlateDecode >> stream xÚн 1 àÈ Y|óÚ;©‹‚?à ‚N⤎ŠÎç£ø"®äb/YÄÁÁ>JÚ¤¡¶Ýèu)¢&Õc²jµiã­õɈZ=Ùìq˜ Y’µh¦>&™ÑéxÞ¡ÎG£Ó*¦hɘR. eΘى/É".à‹Ò¬t ÖòÒªª®ôwèð£VûhOé/oé»2C óxŸûB§©’ÙbM•nÕÿ¼æ¥7÷íÝ¥| "çþÃÔ€3Ÿ©ˆàï>à$Á¾$J endstream endobj 1197 0 obj << /Length 250 /Filter /FlateDecode >> stream xÚu1NÃ@E'raiš=ÂÎÀ1IL¨,… á ªˆŠ¤DöѶã¹.·þü ‘(öigöÿ?³Óêølnc›ÙQiÓ›ÚºÔ'Tl²=ÿy¹ÐE£ÅÊ&•—lkÑ\ÙËóëF‹Åõ¹•Z,í¶´ñ6K³NZäa|„ ø 9€à|t5»¡î¡iûH†„ˆÿ…û JòŽbz<„„ë¼rd'¾¥0Õ´ †½(9qp&8 %? cF¿ûi=¶H^†Qèù #tü)„g/pxLkDÏ…3zô¢ÑýŠA endstream endobj 1198 0 obj << /Length 127 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bSS3…C®B.ßÄ1’s¹œ<¹ôÃLŒ¸ô=€¢\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. õÿ Dü?€‹üÃ`ÏÀOY$Ù€$ ;Rþÿ?óÿ¬$X–ËÕ“+ V—Xê endstream endobj 1199 0 obj << /Length 174 /Filter /FlateDecode >> stream xÚÝÊ1 Â@…á )Óä™ èfaµ Än!he!Vji¡h-GɶL2Ž†€à „á+þ7&.擦‰&“R’ÒYã ‘S2–Ós‹jOÆ ZKFe7ô¸?/¨òí’4ª‚šâ#Ú‚J®"n˜ëŒÝ¯Ãê;€þʼQ¯ýºO„7ZB؈U$fMYDÌ@È ÷¢ÏìÞfò‚+‹;|WWŒ endstream endobj 1200 0 obj << /Length 197 /Filter /FlateDecode >> stream xÚ}ÎÁ ‚@àÂ\|çjUÔê$˜A‚:uˆNÕ±CQWõÑöQô /¤mU²%vøÿ?,ãÓÅœlriâç“ÐÙÁz® mògCsºb” Û“ç"[ËY²¡ÇýyAm—ä ‹éà}Ä$¦ì–ÈE'Â}2´FZ_V 7 W?ÓWãei f åßÊÿ¥)X¹ÊP>•Vž*5y«ÔàMg9V/«Îz¬VJAðf9ô'à*Á¾LCx8 endstream endobj 1201 0 obj << /Length 186 /Filter /FlateDecode >> stream xÚ­Ï1 Â@…á aàœÀMˆ! L!he!Vji¡(X™Í£ä–)×ykkéÂ|°³üÅÒþx$‘ 0‰ S9Ä|æ»W<ìO<-Øm$±[Ø–]±”ëåvd7]Í$f—Ë6–hÇE.™÷¾) ýѨÝ—Ñ ÔFïidjBT’Þ6-5Ú†RPRB *úszGªH©v–*RÍ–*ÒïýðÐð¼à5{ò}ý endstream endobj 1202 0 obj << /Length 220 /Filter /FlateDecode >> stream xÚu=NÄ@ …_”b%79Âø ÕHË"‘ * D[n‚–™£å&ì¶Ü"ŠyãmafôY²ŸíyýÅéõ•.õ\O:í/µïô­“wa\òögÇÊëVVƒ´O¬K{Ç´´Ã½~~|m¤]=Ü(³k}fÏ‹ kEÚ¨m&fhÌF ˜í€hÆrá°ž +'Ø2¾©ʉ3Ùq4|PYáÂÙØš0eܦÑé½³súŸÉ5ɧ¥\Ó@ÜñïeÝ'ýXæÆÆreSU¤4¹äQ~MQdÅ endstream endobj 1203 0 obj << /Length 206 /Filter /FlateDecode >> stream xڥϽ Â0ð+Â->‚÷Z+©S¡*ØAÐÉAœÔÑAѹ}´>Š±bð¼$*.b†áBîþ§zíá€:Ô¥VDJQÜ£m„T‘;÷ýËfi†á’T„áTÊf3:Ï; Óùˆ¤:¦•üYc6¦\ƒ®¾›;ƒ¿lhkb¬Ì⹄€™/N-êÄZ6*±¨ñp·—§¹ë™|ZX›?š¼4®ïìõ½>uóÎæs¾’—n—‚«Ý tnìÆÍ N2\àKKv endstream endobj 1204 0 obj << /Length 205 /Filter /FlateDecode >> stream xÚ¿n1 ‡]1œä%oÐó ´¹”ˆÒ)$n@¢S‡Š Z•µ—¾Y…G¸‘!Š±i…ÄÖ _¤Ï²ý³=¾Œ©¡gzpäŸÈ;Ú:üÀ¡Ù¨¹T6{œ´hßhèÑ.D£m—ôõyØ¡¬¦äÐÎèÝQ³ÆvF0à`ø80¸cfṉ̃bè¢9)zA}T$"ÜË'¯S|_QùŸ(·½Ýª(ãM I +ëT÷PG“eyÅ?¿Ñ4dѸYƒ÷z‚Ü1…ó_ñ ° S endstream endobj 1205 0 obj << /Length 220 /Filter /FlateDecode >> stream xÚÐ; Â@à )„isÁJÐùEü"Ãøb=A×Û çaÄS~]¿ endstream endobj 1206 0 obj << /Length 216 /Filter /FlateDecode >> stream xÚu1NÄ0Eÿ*…¥ir„ÌÀ › ¨,-‹D $¨(VT@I‚vã£ù(>–)V¾AÐaYOòØóç¹??½¼ÐV=é´ÿÞϼÉz`±ÕþìçæéU6£ø]âoX?ÞêÇûç‹øÍÝ•vâ·ºë´}”q«¨µE XÌX™Í¨ÌŽp‹[PÏ0ÔLhB M ‘‡ÀÆ4ì‘™æò±À¸þEâ ŒÆþS“«D¸ÌiDf( šD“œE‹T³HIc %)>—/Ð~Å’\r/_})oG endstream endobj 1207 0 obj << /Length 164 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bSs…C®B.c3 ÌI$çr9yré‡+›qé{E¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]ÿ300°ÿ?ÀÀ ÿÿC=ˆøÿÿ„`üÏÿùˆøÁþ€ýcf ‚¨ÿÿÿÿ?€F€%ˆ5…Æ„ýÿÿ@.ý‡N€%¸\=¹¹CStò endstream endobj 1208 0 obj << /Length 275 /Filter /FlateDecode >> stream xÚ…=NÄ@ …¥ÉMŽ_òÃ(‚†‘–E"T+* ¤A·ÚDâ \%7!H9Ec{·BHLñidû=¿ßŸRI'tT×äò%=Vø‚¾–jIM}h=<ãªÅâŽ|Å•Ô±h¯éíõý ‹ÕÍUX¬iSQyíš É³ã:þ²œ!1¦{.g½‹éì ›t<A9ÀN¤t¿´É½êà`nê [¢Yè˜'ã(3’@øÉ üˆÊ~sPºo£i5¹ÝE,b”³6ÂyÔ0ɬ1$ÄV¸ ç îÁ˜ÿÁÙº[›ìLzõ #¸òºh»&Û;‚þ¡Ä³$²^MR} ^¶x‹?máÊ endstream endobj 1209 0 obj << /Length 165 /Filter /FlateDecode >> stream xÚÅɱ Â@ à: Yúæ ¼k¹ µ‚7:9ˆ“utPt¾>Z¥pcÁÒ˜(¸ÔÍÁ@>þ?1ét>C1¯I0I±ŒàFº–*áx†Ü‚Ú¡‰A­ø Ê®ñv½Ÿ@å›F  ÜG¨` t>à¡ö»îåè'C/fH=û b‰¨ú賚­'b6l öÁ˜í¶ÿÑQã¨"òDõÐ÷–¶ði—¶ endstream endobj 1210 0 obj << /Length 137 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bCSs…C®B.cc ßÄI$çr9yré‡+sé{E¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]˜øÀù(B¬Ž`ÿ¨­þÿ ÂD00 ¢þÿÿÿ ÿaDœ€Hp¹zrrȧYA endstream endobj 1211 0 obj << /Length 217 /Filter /FlateDecode >> stream xڭνnÂ0p£H·ä¸'À ¤Q™"•š ¦ˆ‰vìP+ŽÄ‹eëkdëšÑU‡ÿÇGkÉ?é>í4ëž8æ^¸iÆ¿%ôIi?Ä1B–4,ȾrÚ'û²d‹ ¯W›w²Ã鈲cž'/¨³kL8âïëTó¶E‚ÑÅòÆkÕä%t:u€­=|ðº?õQ ;D»ñN÷ üd~UôÈ7úå ³²S[Øv0ؼ?½b¶j®vÊ?£ ¶kµ1Nš\*ïÎÖ7V§*=4£#SãŒ÷ endstream endobj 1212 0 obj << /Length 123 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0b#S3…C®B.c3 ßÄI$çr9yré‡+›qé{E¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]êÿ``¨ÿÿƒá?œ¨‡ ŒÃ—¨ÿÿÿÿ0Äÿ?€ „—«'W íâg• endstream endobj 1213 0 obj << /Length 161 /Filter /FlateDecode >> stream xÚ31Ó³´P0C …C®B.sˆD"9—ËÉ“K?\ÁÄœKß(Ê¥ïé«PRTšÊ¥ïà¬`È¥ï¢m¨`Ëåé¢ÀÀøƒñC}ÿþ? ÿïÿ“ÿÇðÿÿûÿ òÿÿY–o`*á?Àþƒÿü„Ø!*9 °þ=þÿg„ÿÿÕ!Œ‰@d¹\=¹¹ªˆ÷ endstream endobj 1214 0 obj << /Length 159 /Filter /FlateDecode >> stream xÚÅÉ1 Â@б¦Éœ¸»a­1‚[ZYˆ•ZZ(ZoŽ¶Gɶ 2΢]àÀ<þŸ±óérAšrY;#«ébðŽ6uj ç–ÕlŽj#WTnKÏÇ늪ܭȠªèhHŸÐUE‹€[îÅ7³(Sÿô‹#“d5"${‹ÝÀö?zn<×Ì‘9 ý~qíp%8} endstream endobj 1215 0 obj << /Length 198 /Filter /FlateDecode >> stream xÚ1‚@E¿¡ ™†#0Ðe‰V$Š‰&ZY+µ´Ðh+{4ŽÂ(- 㲘ØÚ¼âOæÏ›$ͦñ„‡š“1'šOš®§6ŒºÄMŽš¤v§¤V6&U¬ù~{œIÍ7 Ö¤rÞkŽTä ï dR" "/x"oø­ß"x Aa…Ì„¡É,ª ªÒ¢~~Ûæ5ÿ¢µÍo×U9ôõú“ö¸qNÈ©9I§‹Rêï Ý3´,hKí`• endstream endobj 1216 0 obj << /Length 221 /Filter /FlateDecode >> stream xڭбnÂ0àßb¨t À½@›Y"QÈP‰Ntì@³óhyÁc×s U‡.•ððɺ“Ï¿m˧ç ç<æÇqÎÖ²Íy[ÐŽl¡ÕœË[kóIÓš²w¶e ­SV¿òþëðAÙtùÂZñJ­©ž10ô€óU¤QÏ"-D×±×ɯ<Œ´ÃNmA…Q/À%n®:˜¨~ÛDGÿ´ºú9ir2ݘL¤y?ÙRΘ<ÚÂè[˜S|—é\ˆŽŽè³ÝO§÷é¿éOýeêÒ¼¦7úF©W endstream endobj 1217 0 obj << /Length 172 /Filter /FlateDecode >> stream xÚµÎ1 A ÐÔi¼“832ˆVº‚SZYˆ•ZZ(ZÏXYzâÅ#l¹…lÌÛXZäÁOø7è†d¨/ã9C;‹GtV²ibsØ0ó¨Wä,ê™lQû9O—=êl1!Ùæ´–Ê}NÐ)!0„Z¼2ó-ygŽÉg"(.’0P5tÅ·ÔAUɲå+Yü0þÉÀ\%å-n¾Ê§—ø¦YW endstream endobj 1218 0 obj << /Length 218 /Filter /FlateDecode >> stream xÚMŽ1JÄ`…ßb˜ÂÜ`w. ~7»hXW0…àVbµZ * vnâUr”aË!ã›,ˆÍÇð½™Ç”ëó«K-t­gQË -£>Gy—劲p3%ûWÙÔt¹’pK-¡¾Óϯ ›ûk¶úµx’z«X §˜™ý 33䎅£r¤CF40Œ@:bª ˜#µàLÉ‚¼ªÁ‰Y˜õ.¹ŠdÄŒ Ç曶åAîȺ ãlBƒ¼³–{,ªZxËÏŽ`1K{¯ï+æoürSËN~±¡o' endstream endobj 1219 0 obj << /Length 160 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0RÐ5T01P05PH1ä*ä26 ¹†™ä\.'O.ýpcs.} 0—¾§¯BIQi*—¾S€³ï¢m¨`Ëåé¢ÀÀÏò $õÿÿÿ?Äÿ ` ÒÍ#…ø$`'0ƒö üøÄù ì  æÿÿÿÿSÿdÖ.WO®@.’Ø] endstream endobj 1220 0 obj << /Length 159 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0RÐ5T01U0¶TH1ä*ä21 (˜@e’s¹œ<¹ôÃLŒ¸ô=€Â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. õÿ`øÿƒùÿ,dýF Éøƒ}üH¤<˜´’ê00ügüÿ¿á?`¨G"íÿ?’üÿ›²Ìÿ¸\=¹¹kqt endstream endobj 1221 0 obj << /Length 174 /Filter /FlateDecode >> stream xÚ31Ó³´P0P0bScK…C®B.ßÄ1’s¹œ<¹ôÃLŒ¸ô=€¢\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. õÿìÿ7üÿßPÿÿ& ‘eüÀÀü€Jþ``ÀÀ$ÀÈ? ü@²†¿•´cg@%å4*ÉßPƒF²øF2?ü€F2~~€F2ü?€NÖ7 H{ ærõä äóV endstream endobj 1222 0 obj << /Length 195 /Filter /FlateDecode >> stream xÚuν Â@ ðˆƒ¥Ð> stream xÚmν Â0àˆCá¡÷¦Õ(v©à˜AÐÉAœÔÑAѵͣÕ7Q|ÁÅAŒwݤéGr—»œé6³&Ø¢ßt°á&…=>'|äÍz z¦zBQÐvŠÇÃi z0b z„Ë“Øö½óoUú³÷ YU¨X)Õ§-ÈؽÈFÅFç'{»“õÇ…¬yVùJtlÉHƒŸ!²r³&µué]ÅŠ;7ä­ØRš“¹ÌCSQñ‹¦ i¬ÊÓÀìw…HÈÂØÂ>ʳh endstream endobj 1224 0 obj << /Length 237 /Filter /FlateDecode >> stream xÚeαNÃ0à‹2Dº%à{pÒŠ.±TŠD$˜: &ÊÈ‚‰ˆ˜73âEòa+RÅ‘sÔÆ‚ÁßðŸÎ÷—óãÅ)eTÐQ‘S9£“mr|IJÒŒæûÑÝ.kÔk* ÔCŽº¾¤ç§—{ÔË«3ÊQ¯è&§ìëµl [fÛ²cv’ŒÓ¨‡¸ƒh+ÂÞÄì¼÷ R PPÛLm5éÄÄ5“wÛƒQ?Ú‹_"|v“¯ÞÖ‰&ÔŠ*þZý³ ÜIM ê]4ÞO©`9k”—å b{0ý‘ýD>€Ø7Æó¯ñùíkƒ endstream endobj 1225 0 obj << /Length 171 /Filter /FlateDecode >> stream xÚÎ1 Â@…á aàœÀMˆˆ@ Fp A+ ±RK EÛ‰Gó(Á2EÈ:/u ‹ý—ÙýŠ™Í§éB"IìÌIR9Ç|c»#¼¦ÝÇéÊ…g··™Ýº«ßÈãþ¼°+¶K‰Ù•rˆ%:²/%!Ô•¥éI­Dã¯eò±äoKõ²ÊhÐÑ°±Œj#0#0£?Y¦` ¦` ¦`Š]ÚГnS^yÞñÊiô endstream endobj 1226 0 obj << /Length 218 /Filter /FlateDecode >> stream xÚ¥1nƒ@E?¢@š†#ì\ ^ c)‘ìÊ…•*qé"QÒŽÆQ8%ÅŠõ2[$rëæ³Òþÿ~þ¸y.9áœRÎ3.žø#¥OÊÖcÂEé_Þ/T·¤œ•¤_Ü™tûÊß_?gÒõ~Ë)é†O)'oÔ6Œ`ÙPv*;k . ,¢ UPC< ”èzDNø‚ùÆe™{àÊææÓÎ¥ÍÿÂ]—É·’~+|ç¢ 2¢%‚¢ê¥E_†IÖqh×Ò®þ xË endstream endobj 1227 0 obj << /Length 143 /Filter /FlateDecode >> stream xÚ31Ó³´P0P°bSs…C®B.crAɹ\Nž\úá Æ\ú@Q.}O_…’¢ÒT.}§gC.}…hCƒX.OæöÌÅò@lÅõÿ``âz þÃç¸ÿC?’¾Æöÿÿÿ¨‡à?P æs¹zrrìRZö endstream endobj 1228 0 obj << /Length 232 /Filter /FlateDecode >> stream xÚmŽ¿NÃ0‡‘K·xe‹Ÿ'´ 0Y*E"L ˆ‰vdÁÚøÑú(~ªwH‘`¸Oº»ïþ,»óë+ßø•Äò¯.ý¶¥wZt’7šjãõÖ=…'¿è(ÜI•Bï??¾vÖ7¾¥°ñÏ­o^¨ßx¸#€È `Î0Ì#,óŽyB=:F̧˜0¤AÌè.O€=¡ðÌ {Å™sØ2tâÝ÷ 9ÈùF¢štJ´£º:ZëTTwHsͪæT«U‹ù!‹ª,†)b˜"†)3þÚÈtÛÓ#}çwo endstream endobj 1229 0 obj << /Length 239 /Filter /FlateDecode >> stream xÚMбNÃ@ `G"yÉÊv~ö%-aŠÔ‰ H0u@LбCQ»’¸nÑ館Ñ?I}ûL§¯óýúeCú-½”¿c»%H00cŽRb†LèÝ5áÁh†¦šRã"Ì&\/d À/©„솄Ná^J¬+J™¯Êx#jCÿ(Ñïä^ ‡NwŒÚ6d`âNùVø?‰1F3:=ª³0+¸(-ª…¶ø aO"{|lñdy‚ endstream endobj 190 0 obj << /Type /Font /Subtype /Type3 /Name /F18 /FontMatrix [0.011 0 0 0.011 0 0] /FontBBox [ 0 -22 48 63 ] /Resources << /ProcSet [ /PDF /ImageB ] >> /FirstChar 39 /LastChar 125 /Widths 1230 0 R /Encoding 1231 0 R /CharProcs 1232 0 R >> endobj 1230 0 obj [46.98 46.98 46.98 0 0 46.98 0 46.98 46.98 46.98 46.98 46.98 46.98 0 0 0 0 0 0 46.98 0 46.98 46.98 46.98 0 46.98 46.98 46.98 46.98 46.98 46.98 0 46.98 0 46.98 0 0 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 0 0 0 46.98 46.98 0 46.98 0 0 0 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 0 46.98 46.98 46.98 46.98 46.98 46.98 0 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 46.98 ] endobj 1231 0 obj << /Type /Encoding /Differences [39/a39/a40/a41 42/.notdef 44/a44 45/.notdef 46/a46/a47/a48/a49/a50/a51 52/.notdef 58/a58 59/.notdef 60/a60/a61/a62 63/.notdef 64/a64/a65/a66/a67/a68/a69 70/.notdef 71/a71 72/.notdef 73/a73 74/.notdef 76/a76/a77/a78/a79/a80/a81/a82/a83/a84/a85/a86 87/.notdef 90/a90/a91 92/.notdef 93/a93 94/.notdef 97/a97/a98/a99/a100/a101/a102/a103/a104/a105 106/.notdef 107/a107/a108/a109/a110/a111/a112 113/.notdef 114/a114/a115/a116/a117/a118/a119/a120/a121/a122/a123/a124/a125] >> endobj 1232 0 obj << /a39 1176 0 R /a40 1167 0 R /a41 1168 0 R /a44 1177 0 R /a46 1178 0 R /a47 1179 0 R /a48 1226 0 R /a49 1227 0 R /a50 1228 0 R /a51 1229 0 R /a58 1180 0 R /a60 1169 0 R /a61 1181 0 R /a62 1171 0 R /a64 1182 0 R /a65 1183 0 R /a66 1184 0 R /a67 1185 0 R /a68 1186 0 R /a69 1187 0 R /a71 1188 0 R /a73 1189 0 R /a76 1190 0 R /a77 1191 0 R /a78 1192 0 R /a79 1193 0 R /a80 1194 0 R /a81 1195 0 R /a82 1196 0 R /a83 1197 0 R /a84 1198 0 R /a85 1199 0 R /a86 1200 0 R /a90 1201 0 R /a91 1172 0 R /a93 1173 0 R /a97 1202 0 R /a98 1203 0 R /a99 1204 0 R /a100 1205 0 R /a101 1206 0 R /a102 1207 0 R /a103 1208 0 R /a104 1209 0 R /a105 1210 0 R /a107 1211 0 R /a108 1212 0 R /a109 1213 0 R /a110 1214 0 R /a111 1215 0 R /a112 1216 0 R /a114 1217 0 R /a115 1218 0 R /a116 1219 0 R /a117 1220 0 R /a118 1221 0 R /a119 1222 0 R /a120 1223 0 R /a121 1224 0 R /a122 1225 0 R /a123 1174 0 R /a124 1170 0 R /a125 1175 0 R >> endobj 1234 0 obj [556 556 167 278 667 278 278 278 0 278 556 0 722 444 278 278 0 0 0 0 0 0 0 0 0 0 0 0 278 278 278 333 556 556 556 1000 944 278 333 333 556 556 278 333 278 278 556 556 556 556 556 556 556 556 556 556 278 278 556 556 556 444 800 722 667 722 833 667 556 833 833 389 389 722 667 944 833 889 611 889 722 500 667 778 722 1000 778 667 722 333 278 333 556 500 278 500 556 444 556 444 333 500 556 278 278 556 278 833 556 556 556 556 389 389 333 556 500 778 556 500 444] endobj 1235 0 obj << /Length 137 /Filter /FlateDecode >> stream xÚ%Ë; 1Fá;ÕÀ]]þ˜ÄÌøhÇL!he!Vji¡h7­eIY@È5`ûNçæýK¸nn…»å;[ÑÀ.þåöäÁ³>ÃYÖûʬýŸ÷÷Áz8nQuÄ¥>Wö#¨D*…Lµ"Q‰¤C–¤ŠÄ¶5e"š ÑŒwžOü©)B endstream endobj 1236 0 obj << /Length 132 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0S01T0¶P05TH1ä*ä26 (Be’s¹œ<¹ôÃŒ ¹ô=€Â\úž¾ %E¥©\úNÎ @Q…h žX.OÆ 2ìÿìøÿ7ÔÉÿ?ðÏþÿç?õóÿÿøÇþÿÃæÿ~0ü?PÀ`ÏÀåêÉÈG(Ç endstream endobj 1237 0 obj << /Length 192 /Filter /FlateDecode >> stream xÚ…Ž1‚PD‡PlÃØ èÄŠ1‘ÂD+ c¥–m…£q@IAˆû;“WÍÎÎL0›† vÙ xólÎaÌgnäû¢ºEãét¥4'µgß'µT¾áÇýy!•n—ì‘Êøà±{¤> stream xÚ…O; ÂP±lãÜ è{IüÄ* L!he!Vj)¨h-GÉ,-$q̃´ÂT;ß…ÃñL­NuihuéÉ—›V'Ç/2OÅì4Ĭx“®õqžÅÌ7 õÅ$º÷Õ$Mô |€ ¨,G\ WÂ{¡ûFÇ9úé^Ù€"J[|š¼ ¬µÐîrè’YÁ"Ö±4nT?…”pGrjݬc_e*[ù«ËM* endstream endobj 1239 0 obj << /Length 167 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0U0Q0¶T01SH1ä*ä26Š(˜%’s¹œ<¹ôÃŒ¹ô=€¢\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. L@ÌÀß$äAD=ˆø$˜ÿÄÿ€Ä?€Ä ‹³ÃÅíáâÿáâ?Å@âP¢&VVÌŒ.ó.ó.S—áG—;ì&.WO®@.n=Þ endstream endobj 1240 0 obj << /Length 96 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0´TÐ5W02S0µPH1ä*ä2 (˜™Be’s¹œ<¹ôÃ̸ô=€Â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. (\®ž\\&Q# endstream endobj 1241 0 obj << /Length 162 /Filter /FlateDecode >> stream xÚ]± Â0†‡Â->‚ÿ˜ÄK…N…ZÁ ‚N⤎ŠÎú¨>‚c‡bMN8¤>È÷] çy’°ÈáÁü GGbŽÎÂO%ÎT2[0“YFK&¬p»ÞOdªõŽLƒÝS¨AZZFý¢HW 2"ÃòL}¦¾Tß©oþýï»­® ËÐ"І¾Öº?¦ endstream endobj 1242 0 obj << /Length 114 /Filter /FlateDecode >> stream xÚ31Ö3µT0P04WÐ5W01T0µPH1ä*ä22Š(˜™B¥’s¹œ<¹ôÃŒŒ¹ô=€â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. õÿÿüÿÿ†þüa`üè?’›îçrõä ä—5ez endstream endobj 1243 0 obj << /Length 116 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0VÐ5W02W0µPH1ä*ä22 (˜™Bd’s¹œ<¹ôÃŒŒ¹ô=€Â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. õÿÿüÿÿ‚êÿÿc`¨ü¨æ`°›ÿp¹zrrléI endstream endobj 1244 0 obj << /Length 152 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0UÐ5W0¶T0µPH1ä*ä26 (˜™Bd’s¹œ<¹ôÃŒ¹ô=€Â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž.  @ðNü5 ¢DØ{! €?˜8$ØÁÄ Á &> F0ßTt£Ÿ©¸ƒa*ˆ`gàrõä äW:Õ endstream endobj 1245 0 obj << /Length 175 /Filter /FlateDecode >> stream xÚµ± Â0DQXúKä'2Ò† á * D” ¨Ãh%#¤¤Âü#6HáWÜYòóMíÄÈà0žÃp œsº‘µf˜¹Øœ®Tz2{XKfÍ1¿Áãþ¼)·Käd*rdGò”R/¥RA-œ%¡a|¸½ݠЂ´V$‘Q¬ùµñžî†·êÞoÄ×e«ú¿U¿ïG+O;ú‚a endstream endobj 1246 0 obj << /Length 171 /Filter /FlateDecode >> stream xÚµ± Â0EQ Ýù €miCp¢ ” ¨“Ñ…(©0¾ó i~ñϧ{~37õ <& ¸ ~‰³¥9—Jƒ¹Ï“öJu }€s¤7©&¶xÜŸÒõnKºÁÑÂœ(4è^J©øåøqÄ^©.JùNQrŒ?)F#ŒPäëQ1H¢)3RŸ;™Ê;Ù˜J~.؆xCÙˆ?ZÚÓOYbÍ endstream endobj 1247 0 obj << /Length 104 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0UеP0¶TÐ5RH1ä*ä26 (˜A$’s¹œ<¹ôÃŒ¹ô≠ô=}JŠJS¹ôœ ¹ô]¢  b¹<]êÿÿÿÏÄÿа—«'W *› endstream endobj 1248 0 obj << /Length 148 /Filter /FlateDecode >> stream xÚ31Ö3µT0P04U02R06P05TH1ä*ä24Š(YB¥’s¹œ<¹ôà M¸ô=€â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. ü òì?Ô¨ÿ„êÿØÿ‘ÿÃÿ‡¡ ÿ0ü`øÁøƒñóöìøØ7Ô7ügø.`àrõä äj'.ç endstream endobj 1249 0 obj << /Length 171 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0S0W0¶P01VH1ä*ä26Š(›%’s¹œ<¹ôÃŒ ¹ô=€¢\úž¾ %E¥©\úNÎ @Q…h –X.OæöX±ûŽììþ±ø÷Ÿýà¿ÿÇÿûÿüü?ûÿÿðÿÿÿ€ùÿÿÆÿÿêÿ€1ˆ ÉÔ€Ô‚õõ‚Ì™2—} ·p¹zrr«xSº endstream endobj 1250 0 obj << /Length 116 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0V0S01T01QH1ä*ä26ŠE-Àɹ\Nž\úá Ææ\ú@Q.}O_…’¢ÒT.}§gC.}…hCƒX.O† øA-Âþÿÿÿ€øÿ4‚Šv@  Ã¹\=¹¹emH™ endstream endobj 1251 0 obj << /Length 136 /Filter /FlateDecode >> stream xÚ31Ö3µT0P04UÐ54R0² R ¹ ¹ M€Â FÆ0¹ä\.'O.ýpC.} —¾§¯BIQi*—¾S€³‚!—¾‹B´¡‚A,—§‹ƒüûõ?€ðÚÿ‘ÿÃÿ‡áÆŒ?˜?°PààP—«'W ŸÒ,5 endstream endobj 1252 0 obj << /Length 99 /Filter /FlateDecode >> stream xÚ31Ö3µT0P04F †† )†\…\@Ú$l‘IÎåròäÒ pé{€IO_…’¢ÒT.}§g ßE!¨'–ËÓEAžÁ¾¡þÀÿ0XÀ¾AžËÕ“+ ‰;“ endstream endobj 1253 0 obj << /Length 157 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0UÐ5W0¶T0µPH1ä*ä26 (˜™Bd’s¹œ<¹ôÃŒ¹ô=€Â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. ì@ÌÀß#äÁHÌD؈:Q'þ€ˆ@Ì&> f0ñd˜82î>3Ñ dfâ ¸™¢Dp¹zrr@Ä:Õ endstream endobj 1254 0 obj << /Length 107 /Filter /FlateDecode >> stream xÚ31Ö3µT0P04F Æf )†\…\††@¾ˆ –IÎåròäÒW04äÒ÷ sé{ú*”•¦ré;8+E]¢zb¹<]äìêüƒõìäðì:¸\=¹¹{-= endstream endobj 1255 0 obj << /Length 155 /Filter /FlateDecode >> stream xÚ31Ö3µT0P04UÐ54R06P06SH1ä*ä24 (˜XÀä’s¹œ<¹ôà M¸ô=€\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. ü òìÔ€Aûòøð Žöêá´ÿ#ÿ‡ÿÆ ?0`ÿ ÿ þÀÿ†ÿ@¡.WO®@.…8 endstream endobj 1256 0 obj << /Length 110 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0V04S01T06QH1ä*ä26 (Z@d’s¹œ<¹ôÌ͹ô=€Â\úž¾ %E¥©\úNÎ †\ú. ц ±\ž. õÿÿÿÿÄÿ °‘§\®ž\\ºâAŠ endstream endobj 1257 0 obj << /Length 103 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0W04S06W02TH1ä*ä2 (˜B$’s¹œ<¹ôÃŒ,¹ô=L¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]êÿÿÿðÿÿÿ0 âs¹zrrå$~ endstream endobj 1258 0 obj << /Length 117 /Filter /FlateDecode >> stream xÚ31Ö3µT0P°T02W06U05RH1ä*ä22 ()°Lr.—“'—~8P€KßLzú*”•¦ré;8+ré»(D*Äryº(Ø0È1Ôá†úl¸ž;¬c°ÇŠí Èl ärõä äÇ\+ß endstream endobj 1259 0 obj << /Length 247 /Filter /FlateDecode >> stream xÚ5ϱNÄ0 `G"yÉ#Ô/iÕ+…)ÒqHt@‚‰1#¶Ó¥ÖGé#dL¥ª‡ãÐåÇ¿½k.Ûª¨¡‹Žv5µ×ô^ã6+ºjóËÛ'î{´ÏÔth﹌¶ Ÿïß´ûÇ[ªÑ襦êûé4”˜)Á pŒàaYàñ˜Y £„¸QDî+ÿ`|ÔÂ.;™1£‡ràÆ °á§ÄšX6”7 !0Z˜6Œ Ós„I¸1Â{ãá8bþgU3/­BF ‘)„™Ó)sàˆ9rá'Aóì±ÀÞõø„·³…Š endstream endobj 1260 0 obj << /Length 239 /Filter /FlateDecode >> stream xÚ1NÄ0Dg•"Òo|û$Q6ÍZZ‰HPQ *–’‚ÕÒ!ì£ýp!eŠUÌ8âi¾ý=o¶ýÕpíZ·-§uCçŽ|H?Я¶\¼¾Ë~”æÉõƒ4wœJ3Þ»óéóMšýã?¸çε/2"På˜<>Ïå uÁfA@5ãž`cÌO4ês´1dµ1gõÊ®šƒîêɧï:ÙôeÔPø~•KÙœ-ª˺QvõOÔhù9–ŒXÒÀÜ…H$%Ë RM ŸÒZÉlémb– „d·Ùr)}ÙA!·£<Ê/}L~ü endstream endobj 1261 0 obj << /Length 263 /Filter /FlateDecode >> stream xÚuνJÄ@ðYR¦‰oyMr¹ÀÙÜÂy‚)­,ÄJ--í–$baé#ø*Ä€…íÙbÉ8ëGió+þó9/wª]ÊiFÛÍ ªftQà5– sªÊŸÊù®jÌN¨\`v 1fõ!ÝÞÜ]b¶:Ú£³5”Ÿa½&HzЃÐZ]À(°&ÐDv) ÿZðÚEÖµ^mŸV­vjRPÜkYß-ÿ™›À€òB4‡x1+É›²>ß[ÐOBò:@|ÓƒFA:änKã¡ýe’Ì4ÒbÚˆå¯Çqã4¿³Kù…mÂÛ˜¡íåxÚá~ÇøÚ⃌ endstream endobj 1262 0 obj << /Length 191 /Filter /FlateDecode >> stream xÚŒ1 Â@EGR¦É2'pÖa!F0… •…X©¥…¢­ÉÑr”!¥EÈ8 I¥ ûàÏû33;MRŠ(‘oSJb:ÅxEk%GU/˜hvd-š•LÑkºßg4ÙfA’sÚÇ°È \à1×0·2wà˜{(Ÿ¡`‚« ÃUOÂ\+rBZt‚ð%p ¬á#'*=•žJ@« šŸðõÒ¿Ï«F»a;ÂWh—nñ ³ƒI endstream endobj 1263 0 obj << /Length 184 /Filter /FlateDecode >> stream xÚmÉ=‚` à’.žÀ߉1‘ÁD'㤎]…Ä‹‘8p n #¡~ $(}úö­ëL<ŸL²å¸6y6í-<¡Óvf{¶ÝÃÅšÅ\¶(â]Î׊p9% ED‹Ì-Æ4 ð•Óžgö&ëÉ{ô¼øâ!1îå¥qƒú?µ\ÀÜ P˜ùCÁµ#ýA“dZz–4Àu ×,iºÔu8‹q…/ÂaoM endstream endobj 1264 0 obj << /Length 190 /Filter /FlateDecode >> stream xÚ}±‚0†K:˜ÜÂ#pO`iÀ‰1±ƒ‰NÆI4º æ£ðõ®ØîKÿëÝùÓd¹Ê0FM•j\i¼jx@½˜%\îPPGL2P[ê‚2;|=ß7PÅ~¤K<ÑäL‰•s ´Â9×óËy|¥9#l K#‚vÓœ_ó[¹Z²½äC„N Ò_‹¦C£•èFôŒÏ,úa8è—‘[NÔøXT®®þQ­€ü÷âŠÝ endstream endobj 1265 0 obj << /Length 218 /Filter /FlateDecode >> stream xÚÏ1NÃ@й°4¹¬—QY AÂTˆ (‘A‹ÃÍrÁå 3AzšWÌJÿ_¤ãæ”kN|y¹9á‡H/”–v¬¹Iû—û'Zun8-)\Ø™BwÉo¯ïVWg)¬ù6r}GÝšÅ3J•~ ZýôªýT™Mè¥Øa.åˆÊ)¥œ- ™oö̤Å/½ó`t™œÝÿ˜þRôø27ÈäVÖ¯½ifðöƒíh·¾hãÛ`+-·Rû¡ÔÑÒìNç]Ódvg9 endstream endobj 1266 0 obj << /Length 183 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bCSC…C®B.c ßÄI$çr9yré‡+[pé{E¹ô=}JŠJS¹ôœ€|…hCƒX.O…úÿÿþÿÿD|?€bØ0ÿ ÿAD}°ò€ÿÁ&> f0ñH0b!þO ¶ƒn%Ørv¸ƒÀî³?sóˆ?À>û æË `Ÿs¹zrríÇG endstream endobj 1267 0 obj << /Length 147 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0b#SC…C®B.c˜ˆ ’HÎåròäÒW0¶äÒ÷Šré{ú*”•¦ré;8+ù. ц ±\ž. õÿÿÿÿÄÿ Øæ Œ„ † ‚ƒ`|$€lthv›bˆ)ØŒ‡6 ¢Žä£ÿQ Ø.WO®@.ÌŒ‡r endstream endobj 1268 0 obj << /Length 145 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bCSC…C®B.c ßÄI$çr9yré‡+[pé{E¹ô=}JŠJS¹ôœ€|…hCƒX.O…úÿÿÿÿâÿHìó"ˆ Á€ƒø$`@±ØCLÁmQDýÿ ÿ!Ä( ,ÆåêÉÈæxô endstream endobj 1269 0 obj << /Length 227 /Filter /FlateDecode >> stream xÚÐ=NÃ@à±\¬4๬¥PY AÂT(PR$‚ÖÞŽkÍ ¸7eŠU†ÙI"QÒ|Åìß{;—Ý5袥ùŒº½´¸Á°ÐaC]8®<¿ár@ÿHaþVÇè‡;zß~¼¢_Þ_S‹~EO-5kVE*#TòÉPËŽaa¥'\¦BÙƒ°û‰«oè¹Ò\Qéõ4÷pf<á¢`2éß”²Oà$‡Ì˜gãßëíµúD> stream xÚ31Ö3µT0P0b#SC…C®B.c˜ˆ ’HÎåròäÒW0¶äÒ÷Šré{ú*”•¦ré;8+ù. ц ±\ž. õÿþÿùÿŸñÿ?cÀÀ€êÄÿÿÿ±4± Nàô%—«'W žˆ‡ä endstream endobj 1271 0 obj << /Length 108 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bc SC…C®B.crAɹ\Nž\úá Æ\ú@Q.}O_…’¢ÒT.}§g ßE!ÚPÁ –ËÓE¡þÿÿÿÿÿÿà >ÿ†ÁŽ±¹›ËÕ“+ H¨X~ endstream endobj 1272 0 obj << /Length 218 /Filter /FlateDecode >> stream xÚEÏ=nÂ@àE.,MÃvN€m M,ñ#ÅE¤P¥ˆR%)S€B‹9QPr„ø.]¬lÞÛÈ¢ØOš·ÒüLÒÑt¦±Žñ&c&ú•ÈFRf1K~|þÈ<—èMÓ™DÏH%Ê_ôw»û–hþºPÔK}O4þ|©…3EÓµ¦s|–Æ@F öÄAÖ¤ÃØÈHaÀž8pnÀ…\]Ï­GЈ-8¶j<ì\  8hP÷Ãýÿø­žHF¬é–=a…‹,oËÚ>“U.k¹9‰s endstream endobj 1273 0 obj << /Length 123 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bCSC…C®B.cs ßÄI$çr9yré‡+›sé{E¹ô=}JŠJS¹ôœ€|…hCƒX.O…úÿþÿÿ€L€Å˜ŒÁN|Œ?ˆ êÿÿÿÿã?*ûÀåêÉÈé f’ endstream endobj 1274 0 obj << /Length 177 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0b#SC…C®B.c˜ˆ ’HÎåròäÒW0¶äÒ÷Šré{ú*”•¦ré;8+ù. ц ±\ž. õøÿüÿÀ ÿBü`°ÿW$þð‰ü{ª1ˆy Ÿ‘‰ùŒ0¢Ÿñ1Œh†í͇ÄqÑ|¼F¼‡ï™aÄ Ñ𕨠‚l¢è·?`¿!°—«'W ±,ˆ endstream endobj 1275 0 obj << /Length 194 /Filter /FlateDecode >> stream xÚUÏ-Â@à%ˆ&c¸Ì 迨¤”„ P‚$Ž¤u½Ö’[GEÓev›¶ æKÞ1Çî»hÑ8º&nL؃-;CF¹XïÀA_ í>¡ôpŠÇÃi º?!å—&+ŒRå"c¢(ɉ(§N+˜ÆµGÍSroˆ‰›‚W\¯Š‹"­àŠ¬æüÏ ¦+éÕtI…–ðߣmÅ›h5|Ö ¸üˆ‹¢dXB]/†qsøº‰| endstream endobj 1276 0 obj << /Length 170 /Filter /FlateDecode >> stream xÚÅ1 Â@ERÓx„Ìt³Ž)R-Än!he!VÆÒBÑÖä¨9‚¥EØq™Š†Wüßü7sžæe”ÓÄ”Ϩ¶xAæƘ‡æxÆÒ£Ù3šUŒÑø5Ý®÷šr³ ‹¦¢½¥ì€¾"h é`,ò‚T¤'ÀuID ˆ§x¸/„ˆ¶Hÿ ¡øÙ÷®î9 ƒ›Zª¯šëpéq‹o¡lª endstream endobj 1277 0 obj << /Length 174 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bSC…C®B.cs ÌI$çr9yré‡+›sé{E¹ô=}JŠJS¹ôœ€|…hCƒX.O…úÿÿ0üÿÿÿˆø"þ3Åþ70`øH؃þ@‚ýŒ`?€#^¬„ùŠ^°Q`Cƃ-YÉ ²œä fƒ€² Ô$êÿ700€ F"Àb\®ž\\æ„wN endstream endobj 1278 0 obj << /Length 209 /Filter /FlateDecode >> stream xÚÅÐ1nÂ0Æñ/Ê€ô–!ïÔ &HYj‰‚Ô •Ú©CÕ @°Æ9j1CäÇ‹KªÞ ’õìåû{iËŠs.y^,ØV\.x_Љ¬ÕÛœWËûÓîHëšÌ[KæEïÉÔ¯|9_dÖoÏ\ÙðgÁùÕ† ùƃHLd€ pÝLià¡'ÒîAi û?’NIû¬ iïÚ&tZÁéà0÷^gú±È…Ÿ¶X{c¹þ‚Y7‘öÉ01ÖÞñ¿<¶5½Ó ¯ endstream endobj 1279 0 obj << /Length 197 /Filter /FlateDecode >> stream xڕСÂ0à›jrfÐ{Ø::"#a‚‚ ‰€€îÞ e0‰XvtmC‚ùÄßöîOõh˜Ž)¦„Š´¦TÑ^á µ²aLiâOvGÌ ŒÖ¤FscT,èr¾0Ê–S²iNûf‹EN†`æÒY9†»Q‰¶3p‚qNÊNÙ3¼ÿ¶ßO0ïÉn‹ßè¶ ×ÄZ¿’J4½&}þ5tÊò›¦y+™A²ý ½-ؼ+Ô€³Wø2>z endstream endobj 1280 0 obj << /Length 236 /Filter /FlateDecode >> stream xÚu1NÄ@ E½Ú"’›a|˜„$ÕHË"‘ * D”H»$*â£å\!GØ2HQÌw€‰æÉãÿmÿ©«ãæT ©å¨”ºæDJÞsÕ ‰gõ­Ü?ñ¦åx#UÃñmŽí¥¼<¿>rÜ\IÉq+·¥wÜn…˜™åº2ûÐÌÌ4w„C0Mý€¤LúNÔéL”túAø ¨9ÁçÒ„Éa=tC¹6”8y€ÇF¢Ì›Ôa¥OÚ2éý/òaÁ<Ãô&ÄØùE>oùš¿åxv endstream endobj 1281 0 obj << /Length 124 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0b#SC…C®B.c˜ˆ ’HÎåròäÒW0¶äÒ÷Šré{ú*”•¦ré;8+ù. ц ±\ž. õÿÿÿÿÄÿÿ¡êêð@†H0 zÂþÿ(Qÿÿ—ËÕ“+ +òT¬ endstream endobj 1282 0 obj << /Length 167 /Filter /FlateDecode >> stream xÚÕË1‚@…áG(L¦áÌtYY +ÄD ­,Œ•ZZh´†£qŽ@IaGhôf'_ñϬ‹gÉ‚#}SËÎqbùléF.b27§+e™=»˜ÌZ3™bÃûóB&Û.Ù’Éù`9:R‘s)U*µH]JóíØý^‡¿w˜Ÿø¤Ôè¨%ÂH«´RQCôª/ê‰~ú´*hGo8‚˜ endstream endobj 1283 0 obj << /Length 189 /Filter /FlateDecode >> stream xÚeÌ;‚@€á!$Ópæº,‚Š1q ­,Œ•ZZh´.FÇ5¸”\5šo’2ã¹s? ›šqòò98^Ñ}G›|ç»9^0ÈväÈV2#kºßgdÑfAYL{NöELi iÛwÐw?>Í,À¨Ì ÌÊ° ]’ xB˜i ¿´LHäÊ›1VÞL0óJRþa”…¢Vèu¦èZ À¥À-¾òVi endstream endobj 1284 0 obj << /Length 197 /Filter /FlateDecode >> stream xÚϯ ÂPð#†Á)>‚çt»ºËÂœà‚ É &5mÂ.øb_CY°N wíztøo,È¿ðNøìvÓéE‚‚ì69‚æWh .-rZùe¶D/@sL¶@³Ï5šÁ€6ëíMoØ%n}šðÏŸÂ :ƒš–ßæ}v%Ö$@ö—F•´T÷iX°zÒûÓ[õñ¬¿VÎÉ!zyMŽì-¹ß+_ªX=”Ey>JÍ3CN™.°àï{ŒK endstream endobj 1285 0 obj << /Length 192 /Filter /FlateDecode >> stream xÚ­Í= Â@à )Ó䙘ÿ"U F0… •…X©¥…¢mñb ¯a—Ò”)®³‹¨pØùà½)6 GqB¼Q@±O[ªÎSQ6{Ì t—&èN¹E·˜ÑéxÞ¡›ÍÇÄ9§•OÞ‹œªªA â‹î¬ì†q“©ÍÒÚÐð@# ~8 ©¡¸ôŽæÚØ7űÚdzm˜'cÈúðh„¢ü/–ämÙý¢:œ¸À“^[Õ endstream endobj 1286 0 obj << /Length 182 /Filter /FlateDecode >> stream xڥϱ Â@ €á”Y|„æ ¼–¶ƒ j;:9ˆ“::(º¶}´{”{„ŽÒ3‘ÞÒÕ!äŽH–ÎóÅ”ÉÄ”'tIðŽiÎûo•ó ËõÒõ†_Q×[z>^WÔånE¼WtL(>a]Qáœ3-c'4‘aŠÎÓÓ|` ÁBAõž™I=E’zNGþKCö ¬8e  œŽpª¬“‹&È•×5îñ ûÚlÎ endstream endobj 1287 0 obj << /Length 191 /Filter /FlateDecode >> stream xÚmÌ= Â@à Óx„¸ ‰‚Õ‚?` A+ ±RK E[“›™£ä)S,;Îh%Xìûfæùh<¥” }å:exÅ\³T¿:8^pV¢ÝQ>E»’m¹¦ûíqF;ÛÌ)C» }FéËEÜ$ s­´àXBט^H”ȃ©ÁÃ@ž?|be¨®ŸàzY©E—ƒâÿðTZ_Õq×-`öRÅ!a~…ˆƒ„®K<.KÜâj/\ endstream endobj 1288 0 obj << /Length 187 /Filter /FlateDecode >> stream xÚŽ= Â@…g°¦ñ™˜„Ä"•#¸… •…X©¥…¢­ÉÑr”aË€!ãN;±˜æï½GÓY‡®âg!ŸBºR¤³@[]/”òw%ä¯Ü”|³æûíq&?Ý,ØõïÝåLƹ©¿+ðx•ƒ“À—´€"Ò¡@±y‰Rx Œ-¶0ª±éþ~Ð*ž?¢uîmÖ½rç!0±ƒe¥æ] ÔEÓ`ç%ÐÒЖÞ*Åsz endstream endobj 1289 0 obj << /Length 182 /Filter /FlateDecode >> stream xÚŽ1 Â@E¿¤¦Ik—9›°° Än!he!Vji¡h›äh%G°L2ΦÐÖ…}ðgÙ?of§óÇœêÅlS>'t#k5Ñ?œ®”;2{¶–ÌZ§d܆÷ç…L¾]rB¦àCÂñ‘\Á¤"iJzŒDˆÆ=á[5/”ÈjLAOåQ~Ñý‰ß¡@«B_ÕZ¯h4èÊJ—â5¡Î«µ^RMuZ9ÚѲuEJ endstream endobj 1290 0 obj << /Length 193 /Filter /FlateDecode >> stream xڕα‚@ à’.<} L— &Þ`¢“ƒqRG®â›á£øŒ—;[pqÓᾤ½´ý 5)+ÊHñ+•9ís<¡’^&¥|ìŽXLפ*LçÜÅÔ,èr¾0­—S⺡MNÙMC±€Ä  ÿ$z1Ú1Þwxï!"Ëûâ>ô<æôZ™iá&³N°?â>cíH ãRa¸ÊÉHŽ'c Ë:ÇÑ´m™¸O,Î ®ð —ºYK endstream endobj 1291 0 obj << /Length 201 /Filter /FlateDecode >> stream xÚmŽ±ŠÂPEï’âÁ4ù„ÌìKˆ¬® ›BÐÊB¬Ôr‹mM>í}ÊûËâì}VÌ™;ܹ“ú³™i©“Ô¥ÖS=Tò'uÃù9&aÿ+óNüFëFü·â»¥žO—£øùêK+ñ ÝVZî¤[(²€ÂÐÛ f#2³;܃J>ÂPD´Cˆv@Z }•ˆ„‹÷c½C  ¤7¸¾Ð'Ð* 4u‘ö.æ7ú¹mp Ìb2ræcÀòÝÉZþI÷_þ endstream endobj 1292 0 obj << /Length 154 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0asSC…C®B.cßÄ1’s¹œ<¹ôÃŒ¹ô=€¢\úž¾ %E¥©\úNÎ @¾‹B´¡‚A,—§‹ÿû@âÿÆÿÿ˜AûŸz ñHð?°*;&põÿÿÿš4A€Åðk£aÿÿÿ[~ `1.WO®@.òÅ^£ endstream endobj 1293 0 obj << /Length 253 /Filter /FlateDecode >> stream xÚ}±JÄ@†ÿ#E`š}!óšÄä”k.pž` A+ ±RK E»#›ÎÇðUò(y„”[,g‚²ìÇ°óÿÿÌÖÕÉzßòq¹áºâꜟJz¥º`;볟Öã íZÊï¸.(¿ÒwÊÛk~ûx¦|wsÁ%å{¾/¹x vÏ’€4¸ˆlnfxYé•DdöItÁ§S¶n\Å#7@efd=º`’El6X4jB*²`„éá¾fÀ}E_éh0‡íb•ôj“1SLÍ€,xÝ>v*‹Å!*:MÃö–Æ¢ó½:²?-y‰%Û§F‚Í@—-ÝÒ7ãè‚> endstream endobj 1294 0 obj << /Length 161 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bcSC…C®B.ßÄ1’s¹œ<¹ôÃL ¹ô=€¢\úž¾ %E¥©\úNÎ @¾‹B4Pe,—§‹Bý øÿ¬“Œ‘ò@dý ùóÿ? ùûÿ ùB~°o’äAdƒü ÉÀ$ÿÉ?Häz“õÿøÿÿÇÿÿIˆ8—«'W ƒzú endstream endobj 1295 0 obj << /Length 132 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bcKS#…C®B.cC ßÄI$çr9yré‡+ré{E¹ô=}JŠJS¹ôœ€¢. Ñ@-±\ž. ì ò ØþÃÄ@òx@ýÿ@ü€á?×C1;}pýÿÿþÿÿÿ†A|.WO®@.üØO) endstream endobj 1296 0 obj << /Length 169 /Filter /FlateDecode >> stream xÚÍ= Â@…_°¦Ð#d. ›ÍŸ B Fp !Vb¥–Š¶nŽ–£xK q\‘`eïÀW¼ïñŠ‰£~2â€cîé!Gš“·š¦ÎO¤j‰Ô .»m÷Oñë1üêâþdXˆ÷„ÈVîŽ|¹¢-M -è§úX endstream endobj 1297 0 obj << /Length 198 /Filter /FlateDecode >> stream xÚÌ;‚@à%$Ópçò.¨H)L´²0Vji¡ÑV¸‰Wá(xŒ…[Æ_­Å~Éü³ó‡Á0ŠÑEŸ_ècäáƃ=’¹2Êb½ƒ4gA ΄Spò)§-8él„ôŒs˜ÃQ¹yÀ endstream endobj 1298 0 obj << /Length 115 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0b e¨bÈUÈel䃹 ‰ä\.'O.ýpc.} (—¾§¯BIQi*—¾S€³ï¢m¨`Ëåé¢PÿÿÃÿÿ‰zÁÀ<Œˆúÿÿÿ7ñÿ,ÆåêÉÈî{\W endstream endobj 1299 0 obj << /Length 171 /Filter /FlateDecode >> stream xÚ½Š= Â@…·[˜&GÈ\@7!Q°1#¸… •…X©¥…¢õ^,7ðæ[n±ì8šÎȃ÷WÃÑ3ä‚r„Å9œAl&’ø]ö'¨-˜\À,¤c—x½ÜŽ`êÕ s0 nå¹Û =œî=Cê¿bq䙣Ò1 S¥e¬”ö‰K•vI'ì’ö‡mrÿ/)Tžòì8R`ßû¾‡¹…5¼ízfÊ endstream endobj 1300 0 obj << /Length 155 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bcc3…C®B.ßÄ1’s¹œ<¹ôÃL ¹ô=€¢\úž¾ %E¥©\úNÎ @Q…h ÊX.O…úòþÿ¨ÿ$þÿ$ÿÿÏÀPÿD2þÿ`ß$ȃÈù@’Hþ“Èô&ëÿ?:ñÿÿÿÿ7 “q.WO®@.‹£ll endstream endobj 1301 0 obj << /Length 183 /Filter /FlateDecode >> stream xÚ}Ž=‚@…‡XLÃvNàBL¬H·0ÑÊÂX©¥…F[Ù£íQ8¥…a†‚Îb^2ï}¹™KJ)*%³ K†w4÷Ò‹ó +‹ú@¦@½á)j»¥çãuE]íV”¡®é˜QzB[Ä_P¥ ¢:˜…ðá9o’.êAµ@9(¡dq%Ÿ»7@â'a¸ý/=ßµÓGÃ.^¬ÄTyhÆ ‰”pÁ A!\\[Üã>P: endstream endobj 1302 0 obj << /Length 200 /Filter /FlateDecode >> stream xÚ¥= Â@…g°¦ñ™ èfI"¦üSZYˆ•ZZ(ښͣä[.(w“€–‚S|Åæ½7q4HRYs_8Ö ù éL‘WCNâvµ?Ñ$#µá(%µp:©lÉ×ËíHj²š²&5ã­æpGÙŒs” V,ÈS*7;(& A‰]ƒt,¾à -À•ÇýGTÎÀµ@Û8×=ÓF–>¼®á ¡¯†¾$Úñ¼Ë_È¥÷ªùF­Ñ<£5½Þ¯ì endstream endobj 1303 0 obj << /Length 211 /Filter /FlateDecode >> stream xÚ­Ï= Â@à‘ ÓäÎ4 ÙˆVÀ‚Vb¥–Š¶&7ðJ{¯à Lig³ Z 6_ñBÞ¼Õq;éQH1µ¢.é„â­#Ü¡Ž$ )ѯO«-ö3 æ¤# Æ’cMè°?n0èO$éòÓ³!W© É¾Èùb Á|3à1³õP¢_6Äæ¬ri©Ölxz+=Õ>jO=®Ù]qÝu¿ôìªÊç÷B·V–ŸÅ´~…º[ëÎÿ)×DÅ\|kse8Ã'á·vG endstream endobj 1304 0 obj << /Length 158 /Filter /FlateDecode >> stream xÚ­É1 Â@ПJø—ðŸÀÝu£Äj!Fp A+ ±RKAEëõh9J¼AÊÁqc!Ú[̃™Ií`4-ØԈËÞð™m»îjw쎜{Vk±«y\Yù…\/·«|9ê½e_Hx’+5ÐCôÑ8´äÂ#‚$ÒRC®¡¹šˆ\õ¡ì¸ÿBÿ"¨¿xo<ó¼âõõIw endstream endobj 1305 0 obj << /Length 185 /Filter /FlateDecode >> stream xÚMË1 Â@ЋÀ4!s7q5Æ@T0… •…X©¥EÁÊÍÑrr‹ñ,,Þ2³óÿÔŽg©D’€MÅ&rŽùÆv‚=ê×þpºr^°Ù‹°Yã—M±‘Çýya“o³YÊ!–èÈÅRÈùr¨êGB®ù7 }Kïÿ´D#"×eZS¨¡W¡ÿ!§ˆ("P÷B Ca÷£}­¢9ª6A«ª=> stream xÚ31Ö3µT0P0bc 3…C®B.cS ßÄI$çr9yré‡+›ré{E¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]ä€Àž¢þÿÿÿ @ü¿A€ÅH2…‚ù`€hàÀ ß €AþAý~ [@óÿ Œÿ€LxÀÀåêÉÈþ:B„ endstream endobj 1307 0 obj << /Length 148 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bcc3…C®B.ßÄ1’s¹œ<¹ôÃL ¹ô=€¢\úž¾ %E¥©\úNÎ @Q…h ÊX.O…úÌÿþÿ`ÿ…¬ÿÁ $0ð()DÚÉ? õþÜÆðêdƒ=˜”ÿH2ÿcÿÏÀåêÉÈÄ£d> endstream endobj 1308 0 obj << /Length 186 /Filter /FlateDecode >> stream xÚ5Í= Â0ÀñW:oéúN`ú¥ÐÅB­`A'qRGE7©…^Ì­×è êØ¡4¾Ø”É? ‰Âé,&žQ@áœÎ>Þ0ÔÍÓ[}pºb*Qì)ŒQ¬¹¢zÜŸévI>ŠŒ>yG”½•¥:ÅôJ•^ý›]ƒS |Á-,ZHZX:È^<rœ[CÂ×Á准’qÊz¤b&Õg¤aì¦QŒ¥À½†¿À•Äþ$›Lã endstream endobj 1309 0 obj << /Length 174 /Filter /FlateDecode >> stream xÚ31Ö3µT0P0bcc3…C®B.ßÄ1’s¹œ<¹ôÃL ¹ô=€¢\úž¾ %E¥©\úNÎ @Q…h ÊX.O…úÿ `Ôðÿ?ÃÙaCÄÙ00~ @2?ÀDv`²N2~¨+þߎ ¿#Èß``’ ?Ÿ‡“¿¿G#«¾g``¨?øA6 Hû†@Rž¡†ËÕ“+ Ém¢ endstream endobj 1310 0 obj << /Length 202 /Filter /FlateDecode >> stream xÚEŒ; ÂPEoH!Lãœø£‚UÀ˜BÐÊB¬ÔÒBÑN!…Û²³t î@Ë!ãL@,ÞaæÌ»·µ{¸£¯Ûá¨ÏÛ™ lµÃfOÄܒ£¹©ZrÉŒOÇóŽÜp>âܘW!kJÆ‹/ŸLnRüQ;”H¡(Ô+€Øû­Üp{Íçh¼¯€/ O ¨.†êçê«oŸk> ¹¶´¬4¶ú…¥4Wè¬&F&ž”™äRŠ¢ª§ÚÑ$¡}¨xY& endstream endobj 1311 0 obj << /Length 237 /Filter /FlateDecode >> stream xÚEαjÃ@ àßdˆ‚ÁzöìØ)ÍCšB=Ò©CÉ”dÌÐÒnÆvÈÐ×jé‹:tÍ&É=Žûîî$%ñÍpÄ!ø:ºãdÀñ-¯"z¥X£!—Znh’‘yæxDæQâd²¿¿}¬ÉLæ÷‘™òKÄႲ)—Ö³µ[{²v§È­õöð+ïðOPy5À‘ Æ@®²äÌ©¤äUíð·-Gÿ[ùÙ;z¿Êßàµ[*ö‚l”ãŽBÉ;¥v\ɼHer”;åSú¾H‹R §Z88 ¾~íKôÑßÍa{ endstream endobj 1312 0 obj << /Length 176 /Filter /FlateDecode >> stream xÚ}Ž1 ÂP †S2Y<‚9¯Å*B¡Vð ‚N⤎Š®­Gó(ï¤Ï¤c‡|?!?É'ãéœSžèä3>gt#Í”»Õ§+•žÜ^wrëŽ~ÃûóB®Ü.9#Wñ!ãôH¾â"Æ…ôPŒ‚¢x+š—"B I À/ >Š¡€i`˜¦$fà_£…$hŠ¡¨†¢Šj(ª¡D{£{-ÐÊÓŽ~æêb° endstream endobj 1313 0 obj << /Length 203 /Filter /FlateDecode >> stream xÚ= Â@…_°L“#8ÐMLRØðL!he!Vji¡h'š£å({„”!qœ-–6ß²ó`ö}›ÄÃtÌ!'<ˆ8 9ñ1¢ Å© å»äp¦iNfËqJf)c2ùŠo×û‰Ìt=ãˆÌœw‡{ÊçŒÞ@в¶^m ´­…ו„û•W÷¨”x:ô däTLdOñ”€_Öû'¤X`–*ºw]!WÒ¢qµ½z¨‘º9KõUóïÐ"§ }}dà endstream endobj 1314 0 obj << /Length 141 /Filter /FlateDecode >> stream xÚ31Ö3µT0Pac S#…C®B.# ßÄI$çr9yré‡+Ypé{E¹ô=}JŠJS¹ôœ ¹ô]¢  b¹<]Øø XŠí¸ˆÿ7001;×ñ¾Äójä‘Ô®ÿÿÿÁÿÿÿ?À0ˆÏåêÉÈÅFJÜ endstream endobj 1315 0 obj << /Length 222 /Filter /FlateDecode >> stream xÚe1N1Eÿ*…¥i|„Ì ð.›-V Ab $¨(U ¤A›Ý£ù(>BÊÑóÓ„,?kÆÿWíEw¥µ®¸kí.õµ‘i;¯O%/¶ï²$=iÛIºó®¤á^¿>¿ß$­n´‘´ÑçFë6Šx0ڄʬ ˜íÍŽX⌾T†~ÂèËÏ°œfGvÄlŽâgØ×ÎOÈ —˜À<|žðHTGÇ‚+î©¥µ§Ë‡D5ÿWôTŒL3ü*Ù¡¸=·‡2šÿÐþ‚½,·ƒ<Ê8hñ endstream endobj 1316 0 obj << /Length 226 /Filter /FlateDecode >> stream xÚEнNÄ0 ðÿé†J^òñ @ZÚHH•îC¢L ˆ @°Ò>ZåáƧúl·ÀŸDZãTåe}Í9W|Qp•s}ů}PYkP·å|òòN›–Ò#—5¥[ SjïøëóûÒæ~Ë¥?œ?S»c„€Nz¬DÈDF‘â˜Mˆ&4=:4§WâLì• «hLºVÆÚšÄQ—5Aýâ1;Í,òw×Ki üs°Ä™ãÇ…à Îdw;«Ò-¯—y"ŸÍ§\Û¼>¹ÿí[z 3áVc4 endstream endobj 1317 0 obj << /Length 181 /Filter /FlateDecode >> stream xÚ•Ï=‚@à!$Ópæ.¿ bâ&ZY+µ´Ðh £pJŠëL±hë$ó%ó^5YºÌ Š(áÍʺÄxÇT²HN)Î7¬4ª¥ª §¨ô–ž×Uµ[QŒª¦cLÑ uMþÁÄ„B9ÓÌÆ›‹‘ñGÐ3aç(if ãMŽÅ( Œ/½#ì˜`Ëc„÷—V2öOZË¿Z;ý®5îñÜþtý endstream endobj 1318 0 obj << /Length 207 /Filter /FlateDecode >> stream xÚ¥Î= Â@à‹À4{„Ìt³&)!à˜BÐÊB¬ÔÒBÑÖ,x¯’£xË’qFEÐÖæƒÙ}o“¸v)¢„ZŽ’ˆRGk‡;ŒSʱóÚ¬¶ØÏÑÎ)NÑŽeŒ6ŸÐaÜ íOäÐiá(Zb>$Ã\CÈÌßÈÌüǹ.ì5ïªTʺ)ñ7¢ ½œùPÐ €ù\è)'…ߘ'å-,e›ù$9óÒ‘• i«ÌŒþ `¾AƒYÒ Öš G9Îð-²c— endstream endobj 1319 0 obj << /Length 241 /Filter /FlateDecode >> stream xÚmŽ1NÄ0E”"Ò4¹ž @’T––E"Th+ ¤Ø´±æ£ø)S„ ãÍ“ü=3ÿuíEÅ5w|ÞpWsÉ/ ©í5ÔgûýóüF»ªGn{ªn5¦j¸ã÷ÓÇ+U»ûkn¨ÚóSÃõ†=6™Ì@! `dÕHpÑë³Îç³¢˜¢¢Œ°0g0º°¿p ã†\ÏF<'Ÿ"D´MÖbLz[‚Îë€õZj6]*7DEñã?°?(£j”A…LP5ãË GÕÔ¡˜µ(O•Y*GÒ@BRƒæ ›è þ5pI endstream endobj 1320 0 obj << /Length 183 /Filter /FlateDecode >> stream xڕͽ Â0à+Â-¾Þ hÓ NB­`A'qRGEÁÉöÑú(}„Ž¤zW©Eqñ _Èå~3°#ò) ¾¦À';¤Æ#ËI~š×Ïö€¡Cµ"cQÍ8ÊÍé|ºìQ…‹ iT­5ùt]ãÁ‘ Ù'é`œ010%p1ßà ­‚içBÆt*R¦—€t 2;nB)¼û½¢¦•×4㪙_T+~Ѭý‹.œ:\âãM† endstream endobj 1321 0 obj << /Length 213 /Filter /FlateDecode >> stream xÚ}O» Â@œ`q°M>!ûz‰I «€0… •…X©¥…¢­É§åSü„”Áõ²W؈p w»3s3Y:Ê'sÆij˜ó1ºPš»¡{¦~s8Ó´$»å4'»tc²åŠo×û‰ìt=ã„ìœw Ç{*ç Ó(¤Džˆ¼`D:„y#jAÔ BQ»SQ]9h@ø”¢9…׆mðÆ 3/"-PIÿoÓ™n•§ ÕªË×ÙñÍó?|ÉR3{¿¾‡6ÒnÚRûúæ}Z”´¡ëån endstream endobj 1322 0 obj << /Length 245 /Filter /FlateDecode >> stream xÚm1NÄ@ EmÉÍa|HB’b«‘–E"Tˆj¡¤`í&G›ŽkøéHÅü 4ÒÓØ£ñnêóv+¥4rVISJ{!O¿rÝ¢‰²þ~9¼ð®ãâ^ê–‹k´¹ènäíøþÌÅîöR*.öòPIùÈÝ^(Ÿ‰(`)3SÚ˜èç¹1›É+-:%ô8p'?, ó\üú‡%ᔀ^Ê‚úH½"È4Ÿ)ÂM¡ñ©úP¨9%7¹Hiè/üŠ!©¯ Gó«dLºâ!n&{„ÁÈë•|ÚÒöÍ J™MøÞc_u|Ç_ž!r· endstream endobj 149 0 obj << /Type /Font /Subtype /Type3 /Name /F55 /FontMatrix [0.01204 0 0 0.01204 0 0] /FontBBox [ -1 -19 45 58 ] /Resources << /ProcSet [ /PDF /ImageB ] >> /FirstChar 34 /LastChar 126 /Widths 1323 0 R /Encoding 1324 0 R /CharProcs 1325 0 R >> endobj 1323 0 obj [43.59 0 0 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 0 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 0 43.59 43.59 43.59 43.59 43.59 43.59 43.59 0 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 43.59 ] endobj 1324 0 obj << /Type /Encoding /Differences [34/a34 35/.notdef 37/a37/a38/a39/a40/a41/a42/a43/a44/a45/a46/a47/a48/a49/a50/a51/a52/a53/a54/a55/a56/a57/a58/a59/a60/a61/a62/a63/a64/a65/a66/a67/a68/a69/a70/a71/a72/a73 74/.notdef 75/a75/a76/a77/a78/a79/a80/a81/a82/a83/a84/a85/a86/a87 88/.notdef 89/a89/a90/a91/a92/a93/a94/a95 96/.notdef 97/a97/a98/a99/a100/a101/a102/a103/a104/a105/a106/a107/a108/a109/a110/a111/a112/a113/a114/a115/a116/a117/a118/a119/a120/a121/a122/a123/a124/a125/a126] >> endobj 1325 0 obj << /a34 1258 0 R /a37 1259 0 R /a38 1261 0 R /a39 1248 0 R /a40 1237 0 R /a41 1238 0 R /a42 1249 0 R /a43 1250 0 R /a44 1251 0 R /a45 1257 0 R /a46 1252 0 R /a47 1253 0 R /a48 1313 0 R /a49 1314 0 R /a50 1315 0 R /a51 1316 0 R /a52 1317 0 R /a53 1318 0 R /a54 1319 0 R /a55 1320 0 R /a56 1321 0 R /a57 1322 0 R /a58 1254 0 R /a59 1255 0 R /a60 1239 0 R /a61 1256 0 R /a62 1241 0 R /a63 1262 0 R /a64 1260 0 R /a65 1263 0 R /a66 1264 0 R /a67 1265 0 R /a68 1266 0 R /a69 1267 0 R /a70 1268 0 R /a71 1269 0 R /a72 1270 0 R /a73 1271 0 R /a75 1272 0 R /a76 1273 0 R /a77 1274 0 R /a78 1275 0 R /a79 1276 0 R /a80 1277 0 R /a81 1278 0 R /a82 1279 0 R /a83 1280 0 R /a84 1281 0 R /a85 1282 0 R /a86 1283 0 R /a87 1284 0 R /a89 1285 0 R /a90 1286 0 R /a91 1242 0 R /a92 1244 0 R /a93 1243 0 R /a94 1235 0 R /a95 1247 0 R /a97 1287 0 R /a98 1288 0 R /a99 1289 0 R /a100 1290 0 R /a101 1291 0 R /a102 1292 0 R /a103 1293 0 R /a104 1294 0 R /a105 1295 0 R /a106 1296 0 R /a107 1297 0 R /a108 1298 0 R /a109 1299 0 R /a110 1300 0 R /a111 1301 0 R /a112 1302 0 R /a113 1303 0 R /a114 1304 0 R /a115 1305 0 R /a116 1306 0 R /a117 1307 0 R /a118 1308 0 R /a119 1309 0 R /a120 1310 0 R /a121 1311 0 R /a122 1312 0 R /a123 1245 0 R /a124 1240 0 R /a125 1246 0 R /a126 1236 0 R >> endobj 1326 0 obj [278 333 333 556 556 278 333 278 389 556 556 556 556 556 556 556 556 556 556 278 278 556 556 556 444 800 722 667 722 778 611 556 778 833 333 389 722 611 1000 833 834 556 834 722 500 722 778 722 1000 778 722 722 333 389 333 556 500 278 500 556 444 556 444 333 500 556 278 278 556 278 833 556 556 556 556 389 389 333 556 500 778 556 500] endobj 1327 0 obj << /Length 312 /Filter /FlateDecode >> stream xÚ½’=NÄ0…¹ˆ4Mn@æ‰@ªHË"‘ * D” hqŽ–£ä)SXfl³›¬Ð–DŠ>É=ｋ⬮1Çs<-±ºÄ:Ç—Þ¡*y1Ǻô;Ïo°i!{Àª„솗!koñóãë²ÍÝmñ±Àü Ú-*¥šA0"²ªaÆDÊ0S¢^hˆF!M$eD–©™3IùÈ•J%Ôia:K)wµr”9ñÏ4ƒ´ä½~Ï„¸]è®Vz^P“ŠíšÉ䞺c:¯áæ2ÔšýšÆñ;Ÿ×'ýRýµW\±;vžæåœYÍy0tY0þÖÓé¬÷\ùqèW|\øê„[ú.9<4cȇõyáÜDËI®æ3rçÔ`Žäs9PÈiP‡ýô9vŸ ®[¸‡§EôT endstream endobj 1328 0 obj << /Length 140 /Filter /FlateDecode >> stream xÚ37Ô³´T0P0b3s …C®B.3C ßÄI$çr9yré‡+˜ré{E¹ô=}JŠJS¹ôœ€¢. Ñ@-±\ž. öÿÿÿ`‚úÿÿÿ€èÿh˜:˜>û £ôp£™? Òì?hEÓü¨‡Òÿ ÐPu ú>p¹zrrå«Á  endstream endobj 1329 0 obj << /Length 271 /Filter /FlateDecode >> stream xÚÕÒ=N…@ðÙPLÃظóðQmò|&R˜hea¬ÔÒB£-ËM¼ÊÞDŽ@IŒ³@ñ˜ÈÜì&¿Ùìÿ‹ü¬,u¦Ïy¥ÞïõsŽoXì¸ÎBž^ñP¡¹×ÅÍ5nôÇûç šÃí¥ÎÑõC®³G¬ŽÚðp#8jD<=€"jˆZ€èO8êb¾lgŒ ~¾ý)z€d…!à+`H70Ø 5#]Áq§kX€Z qBÚÀ à•@âU?£â6êºX êâvFŸ¨>ñüåDmAÀ¹¶ÞD:l"é7w ¢v ;G4äy‰(QHÝ”ÞMœäyyŽWÞá/ó¢,D endstream endobj 1330 0 obj << /Length 227 /Filter /FlateDecode >> stream xÚÝѱ‚@ Г.~‚ýQ'ÅDŒ“::ht†OãSøGÂyG[IœÜ„Žk¯-ãp8R@#÷DM&t ñ‚Q@þv¯þÃጳÍ–¢ÍÒ­¢IWt»ÞOhfë9…hÚ…ì1M(¶î€Ì[xl%Ö¢ÍÅ‚÷Ù4qöésè‰Uì fû–íŠ—È o 6ãòÓAÛÛÿQÍÓÎß>_ëÒ:¥níCûÒ>µo™ÏEçÄsc 1gkù•XŠœï¹Kßn³ÀÃù®îÓ8\¤¸Á­¬ì endstream endobj 147 0 obj << /Type /Font /Subtype /Type3 /Name /F37 /FontMatrix [0.00697 0 0 0.00697 0 0] /FontBBox [ 3 -2 69 90 ] /Resources << /ProcSet [ /PDF /ImageB ] >> /FirstChar 71 /LastChar 80 /Widths 1331 0 R /Encoding 1332 0 R /CharProcs 1333 0 R >> endobj 1331 0 obj [71.99 0 0 0 0 71.99 71.99 0 0 71.99 ] endobj 1332 0 obj << /Type /Encoding /Differences [71/a71 72/.notdef 76/a76/a77 78/.notdef 80/a80] >> endobj 1333 0 obj << /a71 1327 0 R /a76 1328 0 R /a77 1329 0 R /a80 1330 0 R >> endobj 1334 0 obj [556 556 167 278 611 278 278 278 0 278 556 0 722 444 278 278 0 0 0 0 0 0 0 0 0 0 0 0 278 278 278 333 556 556 556 1000 944 278 333 333 556 556 278 333 278 278 556 556 556 556 556 556 556 556 556 556 278 278 556 556 556 444 800 721 611 722 778 611 557 778 834 332 389 722 611 1000 833 833 556 833 722 500 666 722 723 1000 778 722 722 333 278 333 556 500 278 500 556 444 556 444 333 500 556 278 278 556 278 832 556 556 556 556 389 389 333 556 500 778 556 500 444 333 222 333 556 0 0 0 278 556 556 1000 556 556 278 1000 500 278 1000 0 0 0 0 0 0 556 556 500 500 1000 278 990 389 278 778 0 0 722 0 333 556 556 556 556 222 556 278 800 325 556 556 333 800 278 400 556 361 361 278 556 556 278 278 361 361 556 834 834 834 444 721 721 721 721 721 721 1000 722 611 611 611 611 332 332 332 332 778 833 833 833 833 833 833 556 833 722 722 722 722 722 556 556 500 500 500 500 500 500 722 444 444 444] endobj 1335 0 obj << /Length1 1508 /Length2 4262 /Length3 0 /Length 5004 /Filter /FlateDecode >> stream xÚ­”u\ÓÝÛ€”n]%!HwŒT™Û€áÆèi1@ºÄ! ‚ „’’ÒR"%ø¢¿çyý÷ýìŸûºÏ}®sŸó=üÜ&fbª0ÌM¸ŠꆚV@IP\R’œŸ_ ‡à´WååA=w$$@ éKäüuŒ‹7áè„© ÿL’¨¢àX‚BpNpÔq ( 0Ã@pœ·8@‰˜þœá0…»Á±p˜89€! 8ÀM¸#M.ñSIí€Èý/ swùyÀ±nÇR¡Ÿ–€cGôÀàäF˜ãÅàÇ*ÿV×rG" ¨Ÿåõé/A!Þÿd`P.î88`ˆÁ±è?S-áÿ“3„Ãî¨?©.‚D@UÑŽH8@ò!„› 3Aà NÒ þ+GÃþ”8îÜ/ =s°™¦è?gú š@hœ¹·ËefÿûƒExl%û Uì¤3ðX ý|eN ÿâÿZýý«©a¼|Å@Ò1¼ìñ'%}|yåä@þ§2¡îØã…q¿Èã‡àß±âøـýàPò‘A T1Ä9±*ìI€fnOÑYB5Çêh£òÆ/)‚‡c‘ú®"eVûÅ)ôÔ g¸<¿³»Ý­÷wj­¹F' -xØ/¤øÔ²[m¦zÏ»ð즩«êÿº"ChÜ7ñîIœuÞ›ôõ·Ÿ³M„4ÌIǸ šl=*šRCd嬴Rá¡eRܦç“\ŸDH‡{ž{D<ÞàNV#:º©w§ÛÏ%ËŠ=²Û®éx[ØPÙÃã1Oû/2Íz~‚ÃäŒëñ{ÚÌ»ex!›}¼ðQàÈ:ì°´x²íþ•Å¤Hîð ™ãçÆ«–ëc1ʾ’Äù[ÓR¾ˆð£zͽȶ1Tc‰e@´ÊÙž5²¬gB­Ò’MR!< ýaò9ÕFd›‹çgô’·´§¾Ê,=­ÔµDG™.7XÛ¶ÃQé÷D‹itŒâ9ýØ.3G¶ži=ËXš$Þª-sáNÇ”·ÞÄE˜©Ó]\aAaɹœ½àôé¨I +ÃaÎL‘‚£õ°µ} ZŸ5H¢]Ž€ñ„¦WssV‚ –»:°fK`Íâá5Már~’òÐN/9VJª“ä!yÑÅ#ÎGÒgÎÆ€L(ªñ·Ì“ù?‹"|iª=¿Þ ÿ,È)òFÝbRÅup]®§ÿà¡J§ˆN¶ç¹Lƒ' 9Lï“_Žo}wåôUós6óo—´ðã.ÒM‹`¶Tuˆ|þ„Ýá ›Ùœ†Mq}2Áýð›åçn :± ¸¯žd—ÖnkÙÝ+`­Qí¨öÓäfJQ0nÏ({3«Bü6®òêgä»ÍoŠÍ”×x§’}î&u ¿yWG°,ä ½VÊ¥†w~p1ëA¢ MÁà\Ì4'&š yá©K ÷F¤MG*œZVàãÕxuƒÞX.þ8U½–ðòx·WŸ©¦ÌGª¼¥¨Ì"%°…M<7"ÓrKÒ÷ §®”ífs'KJ€=¶WEU7I‹®—еjY=v”$·>CXˆû0šÁð©ŠwÕ€- 6ë:1Ä "ÁÐöí:sµ’B;ÍVèدÎnºk¿ÏÞ›kSƒ±™›«-y³žÙ¾„‡³ûÐóßSñEóÆ6û9w¥·ÓÔ,2ós*'Ó6{Ș&­v‹ æ ͇»ª†a?>í<'õjËöùíq&•Žy×iê]/‡ògÜÚûûÌØ©jœPŒK)©ÙN%Ã¥3ÂC×h²¾Õº´¸?[¶r®'o¹ažwEðÏm¸oå½òáOá²á9okw}ªn¨¬—i4kGïY|çÆNŠ²A—·×îÍôrj:÷€,ûhô‘Nûmœžb¡VVçÍÕõæ2uÿ3i¢f~û¯}îŲ[3©pábREíƒëƒ‹9ó™GCá,¯6ôÐüÕŠV”µPRcM„bP&ï€òˆ$k÷5sW®æôNy41?RIÍõu­iH>åŽTÆË Ó£û¾úwùšÉ4óª„cŠú‡uÎ…ÇX³ì/ÛvùÑy0 ·¿c•Ý$\ᕈcû‚šìåšùKåíÚB)ŠÀ³KåÔ„’<3ë.ÊŽ}_ÑA¤Öúú ·šx%^­6,« ùëÂìß ô»]̘=ã#­l5tÍÖfGO¬LÁw0`!bÉQ.†ìÇWë/þ~…¯á>Á¼ßK|Æß02¢ÍZ’„ìRäÐÙ©¯†ß×w™ìQÊžÓÏlŸ˜¥¿0’æD \i¸;¿¿ëèÿf#p Лjw$ûM‰5à~)n;æ3»E~¡£‚€ª“âMÑU3 =‰–²HëY©yx‚ŠM’R&Â5/oV÷V.u°:IÉÆ‘]ÕAÞëTWá5ªÜ|:'먂a£¥•è×.9ù6O#!?üÏg U${,X€/­z„8’Z‘üpfzWÍ+ÈÑFÂAV$¹éþL‰òR¶TƒrûðºP,ÉÑ—Ü3zöæƃϙV†ì.Ä {A—ûöñxZ¥GÚÒ¬ºè±Ï—he©ï¶fÂñ2Ô5u×½ËÀÜâ5êJ“&¯µ³¦"Ž˜”z¶‘ËÂj‹.Þc¦»)l„ßëGv¯qäÕÜlÔ&L&ã³~´•Z·³òÔ†~ú¹'oJiu+ã<éVhh󲊎ò̀ݎÌA!|ÇQ®ßô²”Nìeu¸×~YUWäÞAX…/ ”œ4Vk”eAÄ(±ý©WËÍö^¯ÉH9žbÊóü« (×.¾ÄuY¤M”ýƒÞxü¼[%uaÇË¢îÙUçm«„ÒÕí¢lZØ'_ÅN¯‡O÷nñ©¦Ç+.†ùºåmBS—*FÛÀ,^Êm$γ> Ì£ r¢ž÷pFÅð‘q—rПó{^À †Œû‘Ôö® xâI1oÀê~.E KI¹²l/EùA€7y_M_a‘£ÍÀ°„Ĭ6Ýôóh:á‰q¦sN”­3 ßôðÂÐDÇ ‚dÍâU➦Ie–=™÷Àg™?“˯5êû¯ ÞaI}îØAs Š#ØRÞ€¦=0y=ãÔAPÀxÚ)¾Í=r!dXÏ—V«n.;Æ@ßT“6w"¾îÒDãÐSk£–ƈð«1 #P»”‚p˜¤vß³T…CG¿Ž‘Å:Óœ—0 t1’‹ƒÿ‘{ÚWά÷éK}"vß„çtœ _Î>²âbÜ<»Ú€X ÓåªÞdx;$<åè;>å=nSê Ó™v´ñ eZ`to¹ –È(?ª»òU\B$ìótpV^x {SÍäK²v_·h`¯ª*ø9Ó}À=÷=j©wy5#‚üÈ Vו*«ì2"²Bª_*½Oï»Ë¶)@o¿ÊÇÙnÀ¸AI0 ³~ãé¨Z"yVžü^‡³[%½]ÞYoøîú_È㊷-¤éñ‘ý¶H¥•» ŸHã4õ”ö0Å([=ŒAe³Çˆ ø'ëïE»Ü‹ö¶È_åq½œÛˆm;$vxÔÕ² áj1«/ +™ø‚³”ÔП¤°¿kþcvUÚè訫Ô6É«:]£ŸÈ‘Q <£RÆÖé3ÛͯÃÇo˜ø½\×*¿3uj ÷\]ñ,e! =¤,% ‡s%vüÐ[]ÚbzŠ©‘z]®±1£Ûß>ªýI»'ii7#áÖw&º¿OíÆrËr‰×/—ÏQoŸ$lVáð#ðL.'¢`±`jYBÓóƒ)ö³ÓîËê# ÷±áV—’ñ—fè™26ì‘°BQs•#ª5ÃN_çÞn±Q•4¾ÿ*Œ{H»\žIž0UIþ@š XVmåíêu¦=xui£|mqð®FÖ âKjÄv¾o0§î®•×¸‹³ƒýô„óâø‡âå´Wd+`Ú a2ü*Û9[–§Ý!.iÐ+±Í5áÌW*[8=Í_ç¹ð\û}»*?„‚å”ÜÚÓmP€ªA¥‚° ŽÆ‘Ì­2—V¹l„†nâZXDÈÖghæ@à 5äËdž°à0i±Ä”ŸˆBé’¥½Œ¦P[\¤eñ?®w£=or¬çQ Ls" søØØTè§Ë< +·Ñâ‘ÌÈl9Á«S¥´Øí¹C¡åTD`_Z{oR„§¯GEQÔ„”Òˆì F‘…·Â­/oºo¹ÖU?˜dqÒ.Uss ¦IÙÝÚËãõT^» Ÿg"|÷ˆ¨¸õî×¢~õI£Q­îsK%Äc®Ö/F?ÓöÕ×Ϭ¹d:@Õο æ|]¦-Š§[¡”¼×k8ßAj8;´g³¨®àaXúagÌ·×ûqDV6I–GLb+Q@MciŠvä"úlNÔsã>õͯßÁÄ%S+!Ÿfìˆé÷nÜÒ³+í èG]­­ËÑÀ¸Üy¼d[°l}«qš¹-<[Ç>C3¢A¬÷uЪÙŠ¹½Ö¨m=kjÂÉB Ëßð^¢Æg¥‚%_=莞΋jݯµƒúª1Ùz˯G#&|åaŽ|>оì¾4x¯Ðâ /»þ(>D‘Té±X/•_7Ózx׊ÄÿìKY¡ð&2†âF!mi¦YªÖ£-Bš–çW´ÌÀùËCYï¸:â(åÙÉ‚¿ènºvù”&.½-× ”H€^\¸ûz¬ZºÊÀf(=ÒSÝŒÀT6{”ÓmŠ)˜Ò Ή«ŠJ¥GìUf¦5µ„ö5B)¯Z¿¹‚²W¯ØbÓÌq×°VPƒ´åÜ“©2*T¹k ¬¼ |îB§žh~cvóìà)ž!î¥í“÷Qì¾ 1yÍÃåã Gÿq†ÀbcÛT~ÒÇNÚ6æÍ™5…P¿MfyËó@nµ¿£ÿÞöZØ|ÄU1t0_·‡( )Ù¹…žÀn(¥±`¶‡—é Ý}ÎaÆ‹°PÛµÎüÅybùNSÚss±Á,â|ýç°¢aKõ62s©¥I6]–Ý«‰øQ6g­TN…ÈÙjŠ‹W4^öœQS®¸-*TÙÓàѶîÇödª\JŸ"é‘Ý:ÄSa!A ßs …ñØÙ9ÿÄ&Y—5ð§nÚ¸|û½g ïHË‘Èô’Õ%çœ8‚!ý¢äj‡EM»­Q±scñp>Ʋr£­Ÿ´ÆÍØkõ°Xº§,Å)t6…Y$D@ïÆã[)ª3,ž)*Þg"v-î®^wùÐ0£8ÇêNÄÐÖÓ¨¥®ý@ãT¢¯òvÓ@oûjuŠ°}ñÞ‡ÏÇHLŒ™Q#S¢‰;…ïベÜùzð-齶Â" tsMLüW¢¿hWô좩Å¿x‰j›Ùô“o¶!Iž¤v4êó‹åò]ó»M{›Ì¥v`r»ë˜—OÉŸ[”ÊWi9R=óá$ØýhÅ2šö9ÚK5X¢ájÖ¤á¯Æð~ö¬ÞeÎË0ÿnNËñiÛáÇ>§ê`ÒVô¯ëÏ’Mp®Ú H÷÷JïÚØÑDM’‰_Ešhï°ðA¢>:·†7hÈé|’™y‹åîd‚½Î2cÝÇa ˜n[™ Åœ¼‹9rGèzÔŠL@åDÛ}™ÉñÝ«Aõ³wÞEéÈ®)Ó~ˆ—/ ,,Wÿ&5%ùBP*]=‰ðšÛö°®|å—žª>½2¾VuÎúš¨Þ4ibK÷i`d@õÍÆäg¬¹/™CØKDŠݱ Roܘµó 1q.*ÄîélSÄ¥#;îûV8 ×á]æfœ ®|Ú1S‘:²m¢ïó[6»o‚)ß;¬ Hð«jLg[’qc\l–:|û •AÙ endstream endobj 1336 0 obj << /Type /FontDescriptor /FontName /JTQSEL+CMEX10 /Flags 4 /FontBBox [-24 -2960 1454 772] /Ascent 40 /CapHeight 0 /Descent -600 /ItalicAngle 0 /StemV 47 /XHeight 431 /CharSet (/bracketleftBig/bracketleftBigg/bracketleftbigg/bracketleftbt/bracketleftex/bracketlefttp/bracketrightBig/bracketrightBigg/bracketrightbigg/bracketrightbt/bracketrightex/bracketrighttp/floorleftbig/floorrightbig/parenleftBig/parenleftBigg/parenleftbig/parenleftbigg/parenleftbt/parenleftex/parenlefttp/parenrightBig/parenrightBigg/parenrightbig/parenrightbigg/parenrightbt/parenrightex/parenrighttp/radicalBig/radicalbig) /FontFile 1335 0 R >> endobj 1337 0 obj << /Length1 1078 /Length2 2676 /Length3 0 /Length 3392 /Filter /FlateDecode >> stream xÚ­Sy<”kŽDÆR–ªW²¤˜{2FE1d9‘¤1óoÆlf˜¡,Y“-‡”¬É H”"¡I¡Œì[h±%B¤’¤oêœ:çëû÷û½ÿ¼÷u]Ï}_Ï}ßšŠ½£6š@ñ÷QÈ m¤ÒÀØ:º"RSSÃÐA¢-q Ð@!4Ó @¢„±.ÂeS0*›yy3MÌöï"í Ò!<Ž ØâÞ /?G)xd°u4‰8|?á8€~ Ý$èÀH€á€'è‘aðÉD `ðL`RRþ Ýo Ðä›Üð-(d €DKá×ùNþ¦~O¾I"aq¾ßÓÿèÒÿð8_ˆÄþ[Añ¥2 °¥@:ùwéað/s¶ búþÎZ3p$&{‘@@¹K¡‡ú‹€üöA,`1ðÞGòà ™ð»~û~[ïÇÚ8¢wü=פ="3œØT@ü£þ#ÿ‰ù]¢C,À ¡ƒ@ ùBþ÷óÏý·b{Éx "{º(}G§ãØ0þñ#„ 2d ‹ï®C¦0øG~kND ö}¬HÄ.Ž#{yÒqxA‰Œï‚¿9Ô¿¹¿Fú“ÔýÎQ©t ¤1ù³ü‰ëéñq:ðßr€ƒ$Ð$ÿ 3âc¾TÛüWQ>HcBþ8HæoÓ/1¿ž/ŽêÇ ü‚øJ_ˆÌô£’˜~ÿøâƒL¢òô+'§ò÷BÀó˃tð‹Ñç3üÓ?Òüwñó/F¥Ð¿¿¿Ý ‰äg§ã?–÷hÈÇ@" dù1=Þä‚ ÒÖ5´ôùB"õÔ©ÿâ™t:ßá7Ç_«Ÿ1â¯"²@<¬¿‡‚7‰8‘v;ª(xo^ÛÕ5pÑÏM%oúEî»T_‚§QY‡T¼W&s‡+êå}ÚàMReè·6ÆcãŽ-\–)®ˆÍá¸$…¢êtúåÙá±ç%VvhLZ×Óþ¸9¬d\GI?£º³A5¤I<‹0aq$¯öÓó1i-‡d­­Šý…"Cròì¡×[æ+CÒwãgZžÙXÛËûí‹Õh'š‰y.à>ŽÂ܉þíí;£¹]à „åõvqun¬·›b¶F«Í=tä°>Ä+õ>¡ŠÈwÄLÅç~8i' «Ûdñh¨­>¢qžº¿v“rÓˆž÷ŦØó²^ˆçz‚åã•›OŒ:—•"£¿Mˆ(æ/rp>wR6Y¾ ¤Tïþä,Jçàí ®³WªëÅzª7 eötõc?g‹by±º¶Þ–ÍA7bv=®™Iùv†·›sùrtn¤öÖs+¦O3²kæcoþ9áÍzƒíÈÚãPüM4æÞÁi'ýóWÏ|-ئÆÔHvÉm#¦PTE)GÛËV XYú­žÏTc«}ÒÓ`›&®—*aß´wúåæ¡‚/ñ5k÷æŽ ¾ŽüJ<º)Û"},‚ å.¬iìV^ry-: -½à{ë"=Éæ^gmÒ&ÍyùÏÃP]ŒÒ‚\í‡@ OvP)­R½ñÆ’žÃ‹`¸À,–-Ïhtã$8Ý9G{Órºû W)…Ø£GÔ3µ¯ûà|G4RZòyèR(ÞXî“ŒMGǑ؈ç«7—_0W½…{%„3ßYé˜}nâsùÅ8õ6¤7Q|Újè#ç±@PbÍgI‡œ¬àólq¿Áp¾ëü^ÏYR„z˜ð¡¸=%cs¥µªóo¢.Ü»[øСóa\€Eg‰0½n¬PlÓøeÒ—š™sæºtϨºTÞ_˜²ËR½‘©žWËQÇ¶ï˜ byì9ÅJ e7ÚÉæ½>cnÖ8ñ‚]ƒp“nCsŽ‹eúŠYÝ)ñó!ÎÖ¾åù×鯯4˜ nC«¯ ~Q¼O™•’hûRntUxt‚ïž¡õ“Ý÷±;Å'»¦ „ô?¹xõ2¶äHG<×ؘ!ùY6ûƒI¹ÒtÚEYÝ 6û¼ZH‰íûç'Ðù‡ýó×ÝKP+2º,DꈔV83EÌ/Ò,ÀÈ7î1y,é,,÷)êšÜ‹`9¾FŸPé® '¬Ëa%…uÍ!V[¤®WQÒ|¸wŸÂô=ÄâÎ4p‹²E5w?Ö¬_ý°j!\fmšÿ8±K¡ð¡ìosÑ=(3Q¥@9"}pëÓ#¢!g¥[9¼6쨛[›²Ý•©òö‚²…hÎÕiÜòZ1Íœ ÁyÙÍ9B/w 1•$ƒRPCÜòЫAÏF„æX¡‚é7öE¡;§/^²‹*ÜîMvì°›Xu °z7Ì\m¾YòÎ= ÔäÓG¯›6בë:Á ÕâNÆãÏ‚à·º—´ŠË k'‰™8õ>Pé“nt¦šÍìÞØÝk‚ì<òáC-÷á3xäÚ¨BœL¦\‚´MŠ=lÚ>jä‰Ø±b?Þ6•{Ñz:eåq‹ L/¡ècºæ@\á“Ï9ï‹J$Ë<Œe¼|Õã äWÅéË´=–¶ìpOKE}¹‹uÏ%K5àåÑ°§Õ»ý°ØuQíÛd[¹³Î3ÉÞSù¸jEƒ ÄåWÝ¢¶¥=Ô(KÝ¢w9Ò,@8XNxOfi¬Oi(ûD×½WÙÞÜü$C)®‚us}Êa ›imèÆÓ¸î©Hàâ¬æ*ÍŽÖš6õôî´6¹tØ óš%×;ÆFË!¾’Ðkòϧ˜‡ja²Ùª[Îuœ¶ƒÆî$Gõº™Žu•TJEUoºÒ%;˜C{¹q¶Åá¸íRêSúYCWóJLàØûìÔ@gâ¾åZ!¡p8´ÆÃ&;U4©¨íd«ç"õhÛ˜YÚ°Öý81±L¹¡mq4±ÞÖÀû|-M^¿žkQ"î:_yŸs$áÀ²¤ÕNüåçÜg­:éòƒÊsßØçÃg–>¿žcÕ¨¾óì;R÷œ£ç)½e`ésñ3HfZZ˜691DPË ßâ!á9E÷ÖÎæk[y ;Ý#ÃÑÚ`»ÅéS{oÐëºFÝWhÛ:‡ÐnwKˆ¡s·C»_ˆW†£—é²ÂzK„°\]ž¯ÇR‹íÔ¼2<Ñ…b-9ê¢yÛxÚ뎶­BØ'Ud ŽÚ¤ ¨8‡²Ðl¹%;ƼØvQŸâézÒ8wÚe$MúKoU»u˜ IØéâ6’¶ÓK?œöäÉF^gÍ_]ó´IžQ9Z§µp0¬-Yª8GÒ°2ßCÇ9öa…¶”$e-Ÿ°JòÊ#ŸêyŽ»®{>ÕPß'+³u·™ÖÍl –N N˜›„xµ/šÊ^ö÷Ž¿ÛÃf_Uf½; +x$xém¡SÍÞ;‡Ôã3æ³P#Š‘Œš]ÒY9Ë´+Ð,ŽxìÙK­fùˆû&Uè†3”×3Ú‘¢–;Ä;É""náðm¥oŠK1·òí”*û-tõûIÒ™~NKWxnëO}%;?Ùÿ*åÊŠ\¼ÇdýŸ£r"‹3ïÎÔ=Cq©ÄûðÓívríj_hlóí…9?-äm-¿Q¶®j)ÚòèöžGiýæj—·ú;ÌÒ1ñðžþ 6Lj*ùðVÌåŒG—ZO\/ÿAÿtì-?aÛŠâb¼Ô=Ùíø¬ Óææ „‰ì¢{ç¬9›zâUQ†+<)éìÎq–Í9—¨·np-F·„‘ç×ù¨{ü ;‘¥I\Oúš­ÞTƒQÉZzèøŽU‹“NU49~§Êë¬#{Ç™» è´¼ÚÇ>­äÍDï5¬ÓºëôµŒ4Ëiænß¿T"Ô/Õp,$’¹°®<Æ´¸O«p%èõÎé FÆ ,.¬Úi1¸¸¨~áü6;•œ!?šÅžß :›E\ʺ¿Æ˜º­HÓ·iþ•mI®åI,¸–ùCÏg}:£÷|Šzrݵ‚ ë&§‡Þjï:íö5”Z5Ar²¾ÙÊ–7­T{¿òLÃöë4·™{!Q¬ö!—¥¾têƒlÚú­°CʲØýˆžeÿµøÉ:ÇþäèÑ»çf×kŇ¿ŒY9.%]aq…Ë•p?yRƒ ¥Hü ¹Y41Œ£› ePâ]‡FŽÃƒMöaîÂû^ ó¥nç•Š©÷jÅÚ(VÁNAE_çÛ¾±«IÍÎÔòæzJkOvÕõ¤A^cþ©Â¨œtÒw´ÐerËpË"Ù™\ÞûæŒ ÏîtÔxˆÉ³D—•ÎÆÇ7WíöðÐ÷®ØÓïØdb²]ã}P‡õÄƤJ endstream endobj 1338 0 obj << /Type /FontDescriptor /FontName /IGNLSA+CMSY10 /Flags 4 /FontBBox [-29 -960 1116 775] /Ascent 750 /CapHeight 683 /Descent -194 /ItalicAngle -14 /StemV 85 /XHeight 431 /CharSet (/angbracketleft/angbracketright/approxequal/arrowright/element/emptyset/equivalence/mapsto/minusplus/multiply/periodcentered/plusminus/proportional/radical/reflexsubset) /FontFile 1337 0 R >> endobj 1339 0 obj << /Length1 901 /Length2 1499 /Length3 0 /Length 2116 /Filter /FlateDecode >> stream xÚ­SyÐdŒ`f†j ài4)H‹Î$z¦8‹ vžNC˜tÎŒLð ä‚<áÌ–à94áÌØ…0ÿ'ÕÀó@öìú 84ÁOa[Ïçˆ\ˆ'ümV¢(¡‰ýèOD…svKfÀ&ìæ‹°è߀µ'"‰LìV’ÿ‡È!jwv‡Ñ9}¿³ t¶ (˜î.˜á¸+&ûrê™íÅmgçàU?5• w«Ô…ÿ¹ýÄL "çS—±…Ç_\Ú/Óm7 0¥:ÈÅ™þQï‹tÎ]Úwìdødr-îï }IʾkLY™½ð‘Å…æäíXäP‹ æî5^ÑýÞ8©I=Ÿ9äQ\ó±·KMÛ2ð¥ÑB»îÓ*=òzú’žçw©>KkÁ3ew2‘¾ùŸ,Êô³•×N–§6¾³*èÂt–û?Ó¾~}ž{«ãW7V‘6Á¬¦ÑBóõ£OxiÏË~t¯šš"upäÏ'îÏ´j0®…MyŽõ±ä6îÀ¬¨sQê FGG&Çn¾óÌœ“Š\jpÒ…Wö¶›Ñ#«®(:˼Äi—5=}Wb¸æÔ‚°’7œ&|U5•× Oµm“õð¶½ÎkPhLzœŸ¨K0®ivð{?8ã7OçžëU’E·ôVÅ>Ù©ÒQ•J‘VðH—’ž/(SI¯ÃPä=CkÝJŸÃQÛ’éy³tînw«‚»)òè6åÈ¥e4Œx¶7´¨==±–7ö6Nº…^^š¬þüªîtùôgwb‰ŠèË=–^d\ëzßÈE¢4‡ƒ(eH²9]&5K?/«o”…”«T-2™S¯és?XgJ’ÕL)é5B¸IWjåýùmÞ-;'ƽÔû¨“-:‘é® Ü+y#ÁÑR/–“ü[GЙ‹8Z[h^’óEªå*t9£X¥0#–S#?zAì›ÉUuÕeÎ5!óÕ.œYž?”UøbèÖÑ„‡]ûoË€…=á2 /ÉÞÔ §Éô€Šõ=·cÊܽš#6ÜWRiŸÌ%Ám Ô 8+¶$ÜoQekìÜê"4ñ¬”½·œ0ÝO¥Æzß.óú{cæ6­’¸U'ÂK«ò1uÒbP8¢S0ºÉ#ì_â‚þ¹ôÚŠ¥Ÿûß©w‰¦³‡½t“´Ên6y*~ÌM¿x>ç@s%‹Ú¯O6èP½0~è°Ü-^jr Õ[ͪáµB'2÷ÞòWfæIý̼ TSAXƒ^ÇI)²*¬w±Ù•“ÍÕQYTj²o`ƒÇ½á³ñoD›‹†¦š›UÇú,Ü¥ª'Ýuþc}Óò}ìMÇ`Øÿ¶#EÑé†n ëå¦áø“ª>™tX¶õÙbÂY뻡Ê037ãLÚpÈ£=²e"bk?§‘ê«¥œ¿#c5ø´ß¤ž´m2¸Öü«q×JœºSçÕ AHíøÞìÜÝ]+”Õ^†õceýuøîigëM…YÇl®ßúpaÛÍ~Ê á3;Î ZÖ_&YÅZ>ò±yTSõéËüéÒ”î×ÇD9 ÛAªi½ªºœQ±†Ú„VÁ"s(­M!¥i-‰tøÞ‡´–J݉Rû‡íµÉš´1+ëØÁêVO¢Î=·Š†¾c¯L6 ­‹’ï1L=éŸ=Úõ¦£ô"od–Ÿ–{°ñ·Dz_¥dþÒOû{•-±eÍÙKL¦%J…­ušeí£­:§2æÆerVç]ó=ª>¿·;âÖàÐã#µÕ›Uî³'ú—1#»6w®JÊd¯ÆQjm”´"Ø1š1HþÊùÃÅs2='ÛŽû%øth‡¾¸ 3Çå˜ßÏõØY*#,!ì~•^½z¾è2ØZ ¹%¡GÌëOx–ºÕD©Ûö¦Òôxþ%ë‰*r­kŠeI£'BØ»7ø:…í8ò^KÃeÓŠ¦!šZÛÅèÑÛ!ã,ñ³µºæ¯R6»¶¤–ÊmˆéÕLʱ8Õ¿ºL¿°þ4Ɇ©|Ðw²ßåLŠ“”LþxYq4Ýæ?“#ýB endstream endobj 1340 0 obj << /Type /FontDescriptor /FontName /HWTUNG+CMSY7 /Flags 4 /FontBBox [-15 -951 1252 782] /Ascent 750 /CapHeight 683 /Descent -194 /ItalicAngle -14 /StemV 93 /XHeight 431 /CharSet (/asteriskmath/bardbl/element/latticetop/negationslash/plusminus/prime/radical) /FontFile 1339 0 R >> endobj 1341 0 obj << /Length1 1139 /Length2 2670 /Length3 0 /Length 3401 /Filter /FlateDecode >> stream xÚ­Sy<”{G)5DHŠk‰™±/%Ûˆ+bT&KŒ™gÆÃ,Œ‘A)keËYCd¹¹¡ì×R”®¥¤‰¬!kiKÖäêv{o÷óyÿz?Ïïçœï9ßs~ç|ò²¶öªÆxª;hN¥ÐU‘jH}u…A"u5$L^Þ”bé•b†¥ƒúROO°½¤ûèkèêkéÁäƒRð 0¥’½ýè °¦âA °I}xÐ"RØ1îÝŒÉ Âa)€5–m’ÙEpX`OÅA =ð‡èÖ×ò¥c)8 ШdÀŒJÁ’ðJ °¢øÑ=ÔØ•M©Þ4ˆèAM•¾´§ó¿‰Õc @ÉðР/H;âÕ`H$€‡ptÀ$BüË,,)* óÍ÷óþi¾ìqŠìñ(ìáà©Rûv܆ʮ²;ù4õ3¹¹‰dƒ%³é¿íç_8– ‘þŠð#´ŸÀo-YƒxÈü3jIÇ’ œ1…HÄ7äk1@¼-DÇy,ÉüêgoúçÒìq}- ·úÕÆm­ò—‚¾‚¶XˆB?àý7í—è¯6ò‡Íž bN5Édßÿ\~*†¢à¨xˆBÔµ´,† €!ØTêZZ@€Ø2d ƒÝ1\B¥³S¶"ϱ•Hƒ}Y#R]€»Ó°8Ѓyã©þH ‰ù Öø þ¶È¸æá~Þ?%kýü)õ{æ—ˆHüÑÔà…Q ¶ò¿;õ_œtHckòõ&a`ºê?0:ÈøQC À½iT¼ŽþO¿.÷õ#“¿>å±!þ ü;íß+61¡2‚TÙPU×ÓÒÚ@Gyî‘8? ¤Ð¿>¶R¾Ûˆ­+d€8X/gîy­*²8•÷ô7neNbu¼ÍÝ{ÌÞ°¾NRA»•òË;˜µ[é"SÜS2þŸ¤|£êÏÚ=6ÿâŸÚóyêŒÛTz`­f>Ý:`Ò[n9lVðªç‹oµ8={ÙQ|õTþÃ,Vël®­¢ÙñéíC²÷ÎTÜÏ×ÖÁ˜§“N\Œ¸£ !‹Jõ)¾¤yqØ_èjâÖá®°`Ï‹<5*ƒsÔz?áµ<žœ¤ÏÎË»®ìÄ5´ÙxIYÛ{‰o?Zq<Ÿ©°zç~õzؘ[Ñ¡@ŒÒÝ’Çó[>]¿"ߧ¬­Á9‘q ­b÷!-$ÿ!ôÄꙃt4¶.êtï'‡ˆÏ–áð6„ŠFðÌ$ƒç󌣆ž/|̹h–ˆ"v(é{.%F?»´k ïÑZPØ.3°·7ÔÜlÿÛ̲>G*Ú@•_Œ(ñîÁrîª1 Y 7¦¬,®7¬éÝÙš{!âŸHÊ×’‹¹-aWk”S¯ND¦úÉx–óª# ®ýJ·*cãÍ#Õk˜'…ÞD­NïˆB]êÑüú¼.éDzJPï“ÎÂodE¶·ÆëÜfu›ÞVA'}4ç ‘ˆœ8xö¡KÌUŽ"oýÄÑž±¹:ef½¦S4È J3IèzÔtD´¿e©dðxyvÄìî_Ô~1å5íSY þøÆÅ4ÏB]pºqx'?Ëþ}’°8QmŸxƒ¼ñå~Ê`ÿ½‹£À¼BÚÍ9¥ç¥ªÅ:ÚÏw‡~zÜCéìu6_ܘ>&—àýÏmËÙˬžÝý/©mµû%aCPÆÏ¼ç¦ yº¼nr˜oZ¾õyuÁy6‘Kb¡XÏžu³Ä¤æTàxQÉ­—yÃVIoÇL³±‚“ÇŸ¯7õ!Vä› ‘È0”@6#U—smÒ¶|rö´Í¶ mæ)¬„– º¢¥+âÙ†X W„Ì»åšs˜ÅåýGŠ Q‹¬µàØÂѬµÞ¥;œÿ ¡< œó±PmC^5´lV¸¶%šÉözña&)óáSùÈ`_©@çµ¹þ׊°± »Wvc&¸(ñ{}¸l6É× zÞÜ8×zóæwRöj G^Ý=zÁðoæÝ)ªºaOm; kåÖ¯¡sòìŒ3å$?­2Ã[0Wãï-Z4GìöÈ+V'ü}Ǭ¢k.aÂÅÔŸœØ¢tËá@éØlΑùõ þØU ËÝâOrømvKŸ×òy…Z5[,jVÃ?G膅7å-RG-’Û†Ô·=ÆöIƒúï_õÉž!eù©¥ÌÝ}®òÉ—MÓG¦+TÞû‚óñ»œR÷£3d³¦ŽœþPÆ#Ûø+¼Ò‰EêOx°ëNK|ÊÍ8ØËW{û@FÁ˜Æ[ÒƳžY}C¦ÂI:ñØI!*e.Ôt°ñ!¼×'¬HZ^ˆoSf^æckºÖEN(ÔN¹žÜáa* Wñàáß²;¸Ð©éœ/èʧ–4Ì2Œê|{œããhý¯Î¦‹£§8‡3Íèg;ö*ënw7m¡o38k;~þal…YqY@iãpü€GØ2ëy´ÑZ¡²÷ªª¹RaLæ¼€ëÁœÑÈ…Ü*0ò²º]®@Ôœyã¾ý-îb7-vÄ›mŠ*˜VDÙZòþ&©v3inKª ÔÚÙÐÜñ´û´æÌ@ºw€®«Ÿ½éˆÄ‘¥=ý.{ÍŽ§†¢çXo3ö;éo‹Ð•Ói:YqòŒÝãNtÚv~=„IÀÕ?³=úP¤•Å3û– ¤òøß”ŸœÐ:nþíÊg‹ò`·Ì '½Œ^‡?¸n‘´ØÔž—\Rœz7GžOždDV?JlãL|”=ŠHlF eœhˆêˆÙÿ{Ø;|Š*|íw”Œ“ Ç+¿Ã†Ï’߬ˆ©Ý…–Ýd†õó<Ò*͛ɺç> ë3yn§Ý¤up\ïÈ ¼;؃Gƒøíïûêze¦MªÊWOZ˜»Ž£÷ðzaÏ㘈 Ì Òì“’É;}¢3º™Rï©Û^AÆè¼ÕV~¥Çnhwbww‹mÆ«ê»Í#â“;ø[›,Û+íǬù_„ù´Wý¡O©Ñq0KY{^¡Õ0~ÛÑ„ØÒožøÈËÀ+KHšyA(Ýó4]4)ì¼vôÉ[…ͱCý^9/ÖjâvòE}ŒäèàŽ©‡;l®‡Š×Øö¿¯šàtk= ¯œœ1+{áöš¦-cä´a\£³œ‹Üüµ:¥Á(¯óN×>N¿ÃC,dCRªŠk%ÄFÄxŽ° FvÿÍ—zA©Û^;-–oQæÙºyÇ[¢ÖâB3Ãø؃Pæª[QR§TSgi¡ÚÉòµÓƒ²_xÜhÙq?cˆ_ÎóÂÒÀÁS¤ïh£±@ÖY5Úâ(!)öNýš*Çù¦BÔÛC÷Ò¼.Î-oš†ït|&@+ +§N(Êsm4¤19­&Š`‰‰¶¤Axåõ«íÓwZz»` í•·U‰8O£Nˆ|¹$iÝzŠYŒrPO“|wɧ©Ë°YÐ0/-·æ,É\Ê=BŠ ó¶Ò VBÛRÎpÚ5¿Ð$.¶?µ¸S+uRöÈùÕE®Ã°®¹^iž×›˜‡•±?ßëÙetHnÉ8.|³õ±ê˜°æz¥Îç˜"d”RæSd¹î‘˦êuvÔ£=3å‹ËÓMp™º(×?>xòWÂÖãÇÃQ*ªY;Ûj™S§ŒCª†?›ã:èÝ!¾+þ2æÀnÞ•,… FhvZ!MeõvçËÜ=9ð2ãP±@BRKèÝ2u³Ýë^â@bŽ uUFt§—h֧ŇP#›–š-…¦¹%Íš.‘+œ/½§ê::JIT×Ú®'{¤Ä}Ò5:b˜ì:\£ëÒoÍdÜö ¦ªs…Cdëq ³>o8Ã]q° çb'©dŽìCɶÂ8ÙXCkÛ†¸n˜{0wYò³ÉÓ$±&„ uÆL9<ÞåDèh£WÌc”3\mþp7÷ZŸ%¬Ž;.Oþ‘Ûô/GÒ×oˆ}ûû´»ö2ü~…߈ði·Fƒ×Uä@ºðŸ§ž½"Ê{ó‰Q1[?ÝÆFä€+GYS áFë-»“×5ÇÍ>ææ勵“UøÔÅ˺‡Î‹®;<örCC=Ê“+úF­¸èùqt‹yJ×E;ÂÓ|Èø›Ý‹~ⱇì¥Y3çy†¥õ¾ŽJähã£ðV]½Œ ­¾%³Â¾Ó†ÉŸ{ᶗN/b/Í4wr9pÌ®»O 2{–{V[‹6' Ã]öç iå²ÈÒÂŒœX¹pîú¶¹Ø0“šný(¹Ï g+ã”ZmÞ—–Ö,¼Iîçä»"'‰°ÇÜ„º ÆU{ƒ˜øôã•Ò]؃\;ÿ÷·ñÒ’öÁ½‰És·Iª÷wˆ¨„¬ˆßã±L.ÌmDz푂áC08¶ØÀËñÀ©éáJyµÑíYSëæ~<ÎXÐ1ÃkhJßþ­zY¹{¾»³Ë€MÅŠú›_ãÚHöC÷¢Y¥fzÏì…/´sÄYE–} 4^;š"Õòå2½V endstream endobj 1342 0 obj << /Type /FontDescriptor /FontName /KJNCRM+EUEX10 /Flags 4 /FontBBox [-24 -2959 1360 771] /Ascent 0 /CapHeight 0 /Descent 0 /ItalicAngle 0 /StemV 40 /XHeight 431 /CharSet (/bracehtipdownleft/bracehtipdownright/bracehtipupleft/bracehtipupright/braceleftBigg/infinity/integraldisplay/integraltext/producttext/summationdisplay/summationtext) /FontFile 1341 0 R >> endobj 1343 0 obj << /Length1 925 /Length2 1691 /Length3 0 /Length 2311 /Filter /FlateDecode >> stream xÚ­Ry(Àl5gV§ÚW\뺊‹ëM…/Þ²Ù³éi¹Çla¦´Ô(ftCøgEÞц}Nw­§îI=ýðëh˜ÿhæÞZE7™v‘#!?į¬¯ê~;Auè|z[pÒ3ïFÎë¶ñ Žš–®/ÄûUD®{‡U^ÏŠ§ê{Xg»Ž*W×Sq^uz |d |ÕÉKîÇÅ‘¨ÙüdšÓºf6Wâ|ú×ÝdSVÓ*;dˆ<õŸ¦¯=W/ãÑjÒ½G,›ª¹§sŽ582*×°l¦&@™¢)¬Œá) ᶑüÍ·*nI»ºç3.\Z+Ÿu8ÁDnýy­{xýk@A±3æ‰B¶BÛ|ŸH²“¢¥Kk…)[*#j_ØÊ„µ;9ÒÂÕiÒ”C²}b£.5 Sþ_Å[ ™dŠ£­¶>»œ]¸ä@܈¸“ܱñ îHìÔýÜ?´™ÿé/˜ý|~æEŽ®:rX\áËŒ¤)IB#¼äYêI\pÁrݵ±‰ƒk&›ƒò¾× ÇÓETW^ÉfC’_¢”)´ôío˜"¥:»íOK{弊n½ˆie Sºâ›[‹ ­ˆF.²!sÔFýyAY¯jâ\‹Ìå̓y¡N·a±£íÍÅK´ÍÒueJ€j*¥:!4”½œ´»‘ØÞeQVóÉ÷ŒÕu›{¢ù˜Ál¯‹åô‰î<ÏJëpAw)Ë>—Y3MxßT¢Cd û)6Ñ?g´8U–q‘×× TLWú øZ6Ôl*¨lìó¹€$Á#M÷—ÆÐ-£X§Ã'{× s÷ªâ¨4#¯eÞæ¨Ü¿Äž?±Ò•Ÿ»ÙO¹¦XINüœÐè@¢˜Xþé?ã|bbG“D~ÞŸ)?X#ÉÌKë}Òª&ÞQ÷u¼ö¹­Ô%±]}yS'»cæj&Ž‘ÈXb‚J¥mê´lچߩw 7¨ RÀ+™ÒïšR VG¾H[~öã½+¤¹ùíZì6)ñt¶”NªË¤©‘$b4²ÚJC”ZZò§â'ÏöþRyÌÝÇÁw’jGÌTŸ,hJOq•ØвÑ43i×ï89ð6cÜ!¤éÜý’½½µ|6ÊW¦‚O4.C°mµzÛæ­ºâq·¶]§½IË8}ðùtÌ@[‡Ó–æÆRZwÅ÷ FxîVɺL«k¿ÌíŒãµr«ÂÁpÂÛΉ l» Kºÿ½AÜk7ñÑ8•*ë*{C§ÙæL±žš’*³f¡—\Çw+FB‰lTÖÁቨ³—-•4™ë+PjÐ8;~,›PѤ¼gSæÎËž‰ð¼ûNYùQ¥qbV¶ŸQ‡BÓÅ!Ëeòô'Óo’ßš \½;Õm>Þd“®~W›¯Oß^GÏîn {»í'ÜèMÂKŠàòŒ¨E§çÉ›ä¯F-Wh—æ;eN¬¬Šµk\ÚÇûù˜’ àç·ß|3åçð^3=ù>ÓõSiý¡ ?ïêzxù:×/ézñÕãÿ÷ñ1óN~âhÕ‰óì×ÈËÀøå˜R¯ÿn¹Öçµö¤[ò¶0,ÕE õ¶à_½*9~Ý=¼&3OëfyÈ°é¤û`ž=™ |™•è^)ÊRή}6 ¯Œ}Zš-aÛàÐæq³Î¢Ä½G¦ VF͈ú*^y3ðÜtó¥ðämùldäÑá v&¶äadú•ÐüŽwǃ­=üÇ$K.ÖØØäÞJÝ>¶{²­§ãˆ¶,»#o—»w”ÙÅàº"’@^ÁœÝ×ipÈ«¥nêÝ´vR[ËJ·>÷'…õOO.»É^¹ÖÖ/¡ýÜquæ1RÌxSƒ÷úò,‰0þ¨6á'½ÍN< ®#¯^'Çjx™£^ñ†U™)}®é;š¢—öŒ{i= 7º1+w¦'´µ×›DÙ¶©ä<Ð+¶ÁPJ‚t‘XÆ»è}ögeî¬9PÛªÒöº¡áyöT?EAžd–_’L¸àJíƒUü,Às K6ûÞ´Èu}úYÔ²%ÚK3œ«r󸣓ßg»k¬¨£]¯†NðŒçbîÕ+ê•GÐékw¬Ðfäbää µc¿®!œÙ½S'°Œ¿-õU¥×c›¨býýþk\³Ú\Ç'­ö¯ê¼ÙXPÊÃ*ºœñE,ιµââzå±Ö^UR;µoÞxÜwI éIJ™µ²—Wºæ™nŽ‘ü >‘fL endstream endobj 1344 0 obj << /Type /FontDescriptor /FontName /OATESS+EUFM10 /Flags 4 /FontBBox [-26 -224 1055 741] /Ascent 691 /CapHeight 691 /Descent -189 /ItalicAngle 0 /StemV 76 /XHeight 475 /CharSet (/bracketleft/bracketright/colon/equal/minus/parenleft/parenright/plus) /FontFile 1343 0 R >> endobj 1345 0 obj << /Length1 853 /Length2 1428 /Length3 0 /Length 2017 /Filter /FlateDecode >> stream xÚ­Ry<”{'²L››¶WÅÏÚ ³ÐŒl] ©±4z+cæ™ñhæ³1£èV.i±tK²•²Zo7K(ÙÞݨnÚ”i¹Õíóvÿ}?Ï?Ïùžóûžïùžcjè³ÙÊ…-…¨DbEÄí»?u“-°Æ1¦¦Ä”ÀÄ)ìÑÎÎБÀšˆ¶ö6d{kƸKypP1ˆ‚D`Cb˜‹@l*ˆÏDÌŒààÐbŠ B.‚¹a€¥˜O3Ú>$‚YLlbJ >ÚÅäÍ Iä8àÂã¿ébà‰!Q$ÄÆaˆDÀ†Y qaƒŸVï‰pÀö3Ì–F|MEB"1:À¢™t¶áÉQ ž&@{A¨’ÿ‡¨ïÉ©RÆä£ô3Žþ#ÍäÃ<ù—‚i¿/„>+Ú±a)ÿû¬§„ɃY.—ÂgSaÄö%¬0ÀaòÄÐ !ìï[£nÍ4Æ{l¦Ñ)tËÏ+ŸÉù0aDB—GüÍ:]<¿Å¨'"X¶p-D¿¯[¿ë厰lák0E"¦ƒ‘ÀN"€6$ ŒÇ! úDH%1è5‰0ÓK$JQëQxYCx>ŒHÅßÀG0Eƒ8’o(ñ úy±Ã6(ÌûòúŸ¶¸º d; ÀÊš„ª$ÚÙ[!æÊXRJ*™93ÔÚ¯1FA2ˆ…¹ß%`9Ä…gUï/‰u/º[:ÛBÕ•[“F;ßØqyξžCª¼“7½„ýŒ©²Ýù£³G ¢Þ/']ÚåûõÏ=´ì{G#CFs¢/,g¼ÈÙ$‰0šÜ7¾ ®ºóÕ’ªw{ÿ­EÐñ–¼ç­ã…>X7ú#͇†*M["+›ŽÆ‘mÔžÿø 3C?laIš}Q:Štõ¾ß÷ņЪµì\’.œ*Ò*ÈøøïÉÅ`à¥ÒÏÐœ`/ɯ[ÄhvìÎÊ%c…íJxpdté%—µ¡ú$ìpe¬XohŽs¸uˆ«áüõp]zà Ná©%ËŽîOp\º²€a~gÛJU])cQq;«„ÚÝ=ËP´ÿ¥Í×j™h *û#æ9$´ëà·ÒžB®óÃ+‹®®î{@¢Ú6"ïndø ¨Û9r >5xÓ4.»þñÍœ‹Ÿ:R[׶^3ëî÷Õ1îóãËNûÅ™ÁYƒ.A«"»ÞÚC„‰± |ßê›V8½úµ6EcìLÜäשsµ©ÿø’à„š$¡3¥Ä)ÏÂÓ~}ì/¢÷j‹¥WÒ‰—N5\kzRÖË·ü4ÇìuKmà¯Ç¸FwÍôüm¡=×¼ úÃáÕÁ[u[èËfëPb ü°Uc `Îæ  ²f¹Ô¹8ulÍk/|´¿FJç‚qË»êDµ«ç²øé|“±ø þeé«g§ÒnyÉÔ@îÀ…£ú'~¯ŒË¯kí64¸™ zW*Šïø?pÔ"Œ›Ú[»–?ô70zC¾vÁ¸çXÍúíÃÕVÆA×›ô#µ3ÚΚçþ$µ2“´kÞÎnk€ïÚ¿¾È&E¢¢f0µ‘YM˜ýÔ¶­úxP‘úÕ‰‹V6¬Ê>‹eù¯NlŽ¯XU¯tØ5VåéÄ`^:“)o¶¯{Æxœ´ó%ƒPîI?Ï’pÇ?ä´Í‡àNvÀêHK‘§ñVìêOsW¥é¼ßmšP\6ºfªxEº÷Ð4äÖ¶ù1beK?Îi99˜õ›òÑ+“¶8Êf%Ñ{äÞá™õö+c«§Vé잟orkaûòá‘*o¯}¤)JÕ¢‹þÏÊ•¾sBpÙ#K8ÑÏ°]l´nk2Ãq­Ôx"c4 yì©bÑðáÂ^eíë›ìR×íñYkLÿT4 ol0hØtݬÏ2MŸZÖ N¤ÛåãŸñ㖪ݣáRKÉO^ø”ö_¯LfœÕÊ‹,ºoL©7KZ’Òî}!à~CrT •È¥»‡œ©=zKðnK£×À½÷ýpî)ç¤Ô~[Øå"^Üü®.Ò³ ´lèícÝ1ÁÉ‘oó7.Õh8µ}Àé¤ÊnÈdB²ƒªæ^ñ¶ý‘å½6ÍÂpH¬YŸÓîöÉr Õžu—¯ÿ2ÿͶ8Ý–%Áƒ~Ý1ÌkI¾Sâ½Û%oH&ôr=4jºÓ¥]‡@a„~eñ»¬ÛùAÏ©QçV­ÔQ²²4ÏJâ«z¬S¨ìjʺðHeãå–O×jZ^¬°:¤¬ÂºK<Ýù¦×—&¼¨tprÁ<×àãΙ¤¾½Ø³nm&»±\až98Ò«xÝÜþ„Yz8xW°ÎCè׊½Í ò¬òè:íUóvƒŸW÷ÐWâÔMžyÑà²[DeBy¦ö­‘ÔÏÚ‹“¯«Ô0wt froÔÿ¸w.ÆñŽ¹ÿ»67è¾ìѥŞ%ºlûOײַ¸ÖD©Æ#¯÷¥+ á¸]„e¥*¬ý+M•ÙKöÞ{;ëúŽŽÿºný@!ÉUªVä’=Nåm¨¡¨wÌ;° CË«xšI[¤èßµïÔúÜjY|àÙøýÉå9‡ÎˆÒ«\{zšzx©NöbÕ-UËv ÝâUŽzÛ’>ýåÖú endstream endobj 1346 0 obj << /Type /FontDescriptor /FontName /HSNTCT+EUFM7 /Flags 4 /FontBBox [0 -250 1193 750] /Ascent 692 /CapHeight 692 /Descent -189 /ItalicAngle 0 /StemV 82 /XHeight 476 /CharSet (/equal/minus/parenleft/parenright/plus) /FontFile 1345 0 R >> endobj 1347 0 obj << /Length1 824 /Length2 1915 /Length3 0 /Length 2497 /Filter /FlateDecode >> stream xÚ­Ry<”ù7®•JŽ’³ž!D̘qŒ£rF59´$¶1óæ`fˆ#„´­;G$ÑtØ£r,&µS®I®DÚÈÕÊ$DôµûÛׯý÷÷zþyÞŸãýyÞßÜÝÃÈŽH hT¦ ²öyáìQÆ’ÕÑq ƒx&D£:â™ €²´D®´pm  0V&æVhŒ¬°/Œ Ò@8 ÒA€2 $‘À~NÁS©ÀQ|H BTì@ ‰¤C¤ &°ÓAØQ@:DÀS<3¤ˆðdÀƒF€@f$°#“ÜZÀ ²(@„L $ATYäšüÔ@€ù&†…ü é ÑÀNÑFú€h"JŽi ”EºÒD³@‘’ÿ‡¨ïÉÂÈdWÃÑ5ãäóBo¤˜&ž”ÏΔlOˆ N–©Þõj–V¦¸T*s9kõ§uVç;Ï5á(9R Ál™d5nlYÉ-Ö•³ÕŠã;·¶µÉ‘òYG….ñd?”Ql«ªç8‚€š›¶jz'³†m´:ÔU­ó©Úxu‚O‡ê š{]¶1ÃJ,!Ó—¥¡’Ö^zfƒ]îõwàŶ·Ëž†Ž¾ñ+f×o ŒŽþa¯PªEÕaΔIéo'’3¨^Úíø¿idÓ²åÂõžf’g2¡'zãÅ÷{5õ3"tcO;4®Ü]¿çM¿eÅ9­¿~Ð7fBw¹ÇB©ÝSú‡Q#QÚƒL VD³n¿“$¾$@¼=ÒyKÍ°7,?yÙ|»O Wí ¨‹í?.¸¡Ú„˜ù lk,ß°MÌÛ¢%pߌ¿çR¹ýœ[°~{N‚Uë&Ö7<Õ»ljš=#núš¹NG…Œð5ÜݾL3ël¥ñÅM›á?å¨K¿0Iò²ù€¼ó°g.[˜Þ=Åzƒ•çàtC[0¯4_X Þ¯ÜjÛ`–¢Ï5ài&'¶¿®¾Tˆ”!sç#ìyD¼÷>£]Y Úi'…ž–K©>Øî~‘ÍWÑ|v÷Z *æAÕ"AÏ9AäÚU`'ô¸sû×5ôã+râT¶ÜtUÀÞdIœºõÔß¾R¦”[v6/è¸qw"ïè£à­jZX í{J]þôÎ$\÷T‹£÷úö–¨¸Þ‹«ó(šåõo †Fœ{©n¢:Ëq‘^ÐΫ»«Ï­ë“õ%ê@i)ï±~'y¦ÞDx÷~a¹“YpöoݹÕttkIMM‘]Yó•ÔM§Þ’èý»œ}‚ ‘ñ†Ý’Q$':Âq„@ŠZ%{¿©«÷«'-ä/ŽT1y™ûÆn~$Õ&'MCÚΖŽ/Æ Çñ ’s]„ Ä´º§¢%‹õ3ù³í½‡CXâåœÏ’uV'Wm-‚Á…Ì/CµŠNۼωˆï9ÀPΦMÖ\²8—§/È;“[,ØJ/%~ãs™êú£µ‰5e~¦ÉòÙÓ¨ÍU¥X—/;Ä™W7«g=ÑãGë /U_FwÀ¶b-w4?•æí®_YÚÕÞÐ-^“ö°§œØ"ì„{ø4L’M_ÅY‘{rþãÙ¼ëå7;¿[PÞfy{k½LO]/j장3.×7Æ¥ÀgX•±òè™F:EÆrÈÜa½!–÷p]CÕùš¤NTkà]±t+ÈßV©=GF5+Qš¼{žÕ$Uÿ„'s’—4'Gä¡iå½›¢aCØHþØ–#IÛûÌN0|4ê5»0I§::g]èr⦺¨ÁpPQV󇑉r?–=Õëîß´’ç5fâ7Å6Ë;hÔ}¦YrU²0*Çb6äbµObÚ‘þ¼g1÷ú=#r·åú£Ô<ÏYö~ÙçãŒEiÏd\ÅrjÑ@„Þýmå ƒ*…Vm¿Ýz¥ï]²~YÜš!s¿Г¨óÿ•š®xQŠÌý¥R0Þ`Д©Òf’(ÆŠ³9T2|!:î帲üÁ´¨k¨—©?œÞt.sîL´n¾Cer§£òÎ)êa8ÁxÙŠú…NÓõ][ã÷üG­Lm1üÕ<"U³È›ì‰)X œ>|·ÇüÕ[°Jï¦'Ñg ä/ÇÅ&ê—zj\«œË²âëû"?û­W•~®‡é9má=r¨6í‘f׃§ñ˜W¬ÑÚ·³¥ù{¼ùíØå}b]ËÛ\”4:„çýØë^\ënÞÑPŠªdÙÜòÆlYyeehx»sÅ9Gé¬þHíÊR.í´¿Ìfѹ0Zñÿ¸ÀoGœãþyU§¹Oï‚šÜÞ÷3S*Z/Ÿ+£øJ¸µ5Íg>ót;f»ÒÕj6ß´63NlÛ§PO”ÓlÔïîyãj99zæ¬×aÙª™,³Ÿ–>߯µxúéT¸kqTú榥Œ™Ç‡CIÍÚg9ö¸vœ lB×0áFlÊQx2?ζ§Uí›Tæ<|"Dôl8°ÿò~ñ^{eO>ûñ“%ᢀE1zöbÂú3NÂwh:zKÇ}ëp6aÅ!Kj*"¥ž¬²3(þq½£ØexWÎyP§O(h™Oåß‘ä•««ÌV¹X]k8ý2ì°7ÅB[yÂÁÒ¥¼">ðé•Xð‰i‰W¢Ç Ê«¶»»‹rµøä!”^ð?^mNZÏ°pµŸÛsßxÉÉ0ðÏ$ëK‚U—ßy¯<þþÐ͵ endstream endobj 1348 0 obj << /Type /FontDescriptor /FontName /BYCRVB+EURB10 /Flags 4 /FontBBox [-32 -208 1178 723] /Ascent 702 /CapHeight 702 /Descent -189 /ItalicAngle 0 /StemV 114 /XHeight 486 /CharSet (/Lambda/phi/psi/rho) /FontFile 1347 0 R >> endobj 1349 0 obj << /Length1 819 /Length2 1904 /Length3 0 /Length 2486 /Filter /FlateDecode >> stream xÚ­Uy8”{¶§Q#”-ôR’ĘË1””’¬eK ó£™w¦wfûÕ)e ƒ¤ìY ‰lYZ„„,E…RŠrDCgè;ß¹¾ú÷»Þ~÷óÜ¿û¹Ÿç÷\׫ªlc¯iB¤y€4ˆ©‰ÖBãsG;S}£…F¨ªâaÀ$Ó 3Äh `Mó0ÚZ‡ÕÃay$ÀœEa€Ä`~ DAö‚@"àáìa*‚¤Å#ãit˜ìåÍvàÕWõ*“= pˆÀô©¼‚ž `Oó$ƒL-À„BìVn0;¾ Q FD²'ð½ÈµâÞ"ÑýŸa"‹þOÊ„¼€¼ŽÔ^?DDñçy$!PÖ4^-çäÿaêWq …bM òäW'ú[š@%SüÿCX™à¯„£àOG¦4ÊoÚ–L…ìiyQ@@ûgˆÌ° ³A¢ ™éé ¸!⯅y³Z-‹2Å°wqÐøùà«9b:øÓÿ«ºB^Åè1o"0™ ¸jkik£yDÞ÷ÏéØ/µÌ!O‘ y]=€ÃouxHDdˆ²Í3ŒÒ‚hLÞ€Îbóv F¬É•Ñ!æן ïä7õºk]~¯§N,r ŽŸ’ûøà©ÃeN‹78RÈqáñ-~ÜÍŒ?kƒl;,>…ŸŠMé]÷=1Î ¨Þì4Ë9äÿ–®29%^Sùìë]þÃÝÃm‰ÎÙ÷3>?˜Ê²Ùaæ0±fH™¯ÑÕ·¢1í´ž¾“‡âx&ªl;VÙnCÊ©‚³:g^úmHŒzÙâsF´Jcp†VË’\¼.z5aùøZ\L÷ùGvÔ$a Ÿ\Ñ3ò !ÙYš׎&ÊgFtË´·#½RC]fE4!¾dS÷Þ©úZöÊ£ºp¯|·š_ˆ±J×f9Ã@ønÄfOç.¹—˜†|ÿú8œ L¯Îe󴊾åí*M3S2'’•ƒÜ“Ù›p¢Ú2výûtÎ7°þâf,îY” ©"—µF³äÈ“ù¦ó9nu‰yŒbdõÑ«ãçÜËÄôe 1;§Z ÂÅÛ©ª­¾Ïõ·k¨…ŠYž›¸ö1£Â±=Bü´Çª4ÍÏÇíR”ä¤e…”Oß…›-äÆ’öôR´WÚò––Á7FøÛ‰‹æŸÁ‹8l¥97[{‰÷KÆpü£Z¼–Ó6‰¥4©WÏÒ=˜*éÚ@ÂõßZâDg'©Jè ±y”Ÿ~'!%ï«ÔÕî0¿Ý(Öþæ´åýrX¶~“ÓÙr$ü}Ãà«ÅŠhÛãZNñÄ…¤ÃˆÆó…40o½l–ïîÝKÅ=q‰zÌǪO] $];J­N¡ô´{~Ûl.,]†ïÍÁY¯Îß6Ý.œý­„-êMïÚ³±H9IØùV!W°!áà•wWÝZýZ¨"ç6|W@lÚÒ×èÑ:Ù¬8#þ¹‡ñØz£U2<£æ0àxAü@ûÈ– ³6æ²|³·^áöÓ0³0‰óŽ¦}¨#–O²Œõ¿Ü€ù…TƒnêÕï|¡ ·×5½¢iáUªÅ…_1qvά@äM¨\„}fÍ€Û¨¨Vìàò™¼\†]ÁÜtE½SÙGbŽ½u³þÂnAÚDZçòÉ%¥»:ówä—‚R§‹Æ{ÒJT+`˜Ícâ—üæÅ”&¾Þ¡|Gå&Œ¤C»l}tìâºÃWß^ÇöšŽš¹Ä\æ3Æ©Åv(¥ë’féz\mÅ)*âÅÞ6 ä‰,9p¨a Ù æí­‘‰‘ï1,Ÿowj]0½Ôaòøõ³2!ÿ]ÍÎSzcÔäoêùs"îB-6Á ‚³YÙo溤DêÆ>+îæ»MÒ,|»t,Þ`¼l¾Å^5檬Dàã %Ø#‰ø;õw(|Ó¤»¯¸–AÕJ6k:hÏÍ"¤iœ›á©géUÇÛØ®@œás°^I›”é-  ™#h µªûÅñþr&ü_¿ÛÀSÛŠõ¹#u nõÂVëË»˜°0c«¤l†4©×§0„NâˆÜ=¤ïN62Ê4 æúZÌùô&Äcw­…×ÅÈœì÷`d˜Åß²F‹a³+ƒÔ”>kGÍ%æÞé52ûýÂ'¯úfŸœ¦ì3wY¥ˆÁ=93†•š†~çÄêà§#»ÍÔ-œD•ò—>Í…_ïŽá–0ªÙ»#VØÏఆDŽI‰w­¥2ðÍÁøDáG¥Vi¹,»‡»«êÓ:߇IÛŽÅ(|òyýVËTÑæîŒø±ᥩ × QvQî«ýHlÐÝ“:jnñ/œä÷TßèÁÎôa$E¦îºÙÿ0ÜX­®s¤87!€Tò2Ü*!pN}jR, ¼¡Üà]›1öš â ­bïlÔl²´*2¾¸HlŸXoõúÂroï ¼–dWp´òX^Ï2bBL³øò´ê³”ÏƦM§‘ð-Í}§8dÛrzºãW£— Où*±¨<[ËO[•²2ȵ…}7 œyrÜ¡›+Ý~Q#ʼ(´óÄŒø~ã3Í)¦ª5§&ûÙYÑϤ*>,®¶gê*]x‘s>®»m_å&æ_ÞN1‚šCv³^Y÷Öuƒu²n'“í&NVf†áHƒƒÛãÕþJo­c,àŽêr]bŸÜáýŸOúÖ™¨~¯yAÙ|Ä*rPÄU¾_J¾2. µl”ÞfpönXFϘ;r퇤ĉQe§oÛ©`sEœ˜Wc–èçÒdß½Ìlƨ¶L.ÙN”, »uv/®¯OºhÕ•žJ²äún¼2œ__Ñžš *¹!ÆIê ®ë7ì¶ó{ꊰ½ä2šGìöJŸnF¹^9¤¿æË#œŒtM•–a‰síüäƒT©Òç:)³ÍòÍØ?ÌÊÒBÄ> RCÏê;?_‡{åÈ YŒšJù¸ó™ë¤’ª ¯|ø´vW¦Õ½Ô½¬ÖS'ö_×ë˜ìw°Ü§xªÑ+Ê"ÃÇ¿.ä5ý4åëÒ°íÉ!+6a’´;L¤qýô<½à&t‡Âõ&ÍgÉb2K¹)u_.çêIß›~ø ºíÅîwµÁOµœ}1ƒ6Œ’›•]Yñ7¸à endstream endobj 1350 0 obj << /Type /FontDescriptor /FontName /BCJSZT+EURB7 /Flags 4 /FontBBox [0 -250 1329 750] /Ascent 703 /CapHeight 703 /Descent -189 /ItalicAngle 0 /StemV 124 /XHeight 487 /CharSet (/Lambda/phi/psi/rho) /FontFile 1349 0 R >> endobj 1351 0 obj << /Length1 1565 /Length2 11418 /Length3 0 /Length 12345 /Filter /FlateDecode >> stream xÚ­•UXͲ­qwÁ™¸»»»K°gâ:q îîîîw×à!¸w—pø×Ú{%gÛópÃÛU=ªjt=)H”ÕDL쌀’v¶N ,Œ,¼‰Oª ,ÌVF 1ÐÐÉÂÎVÜÐ È `ááa(Ú¹X™,\¼lœ¼lì gk `ú®àp‚€ £…™-Ð`ä‚l mm:†ö¦ŒïÉbvöî 3s'µÍ?Š\ ÈÂØР`èd´y/hlh P³3¶:¹3D¬­ªÿìp¨   # ÀÄÂØ `4³°E`ú§}[S;׿—Mœíÿ7ä9¾O ~Ÿˆð>‰­µû{¦LŠvïµ€ïüÿhê¿Å%­­ mÞåÿméÿ7´±°vÿŸŒ,üïMà¿[RšX8ÛüwTÆÉÐÚÂXÄÖÌ`þ÷’…£¤…ÐDÙÂÉØ`jhíü×:ÐÖä¿K¿Ûõ¯ÂLÒrJ*êtÿsèÿ *ZØ:©»ÛÿGöŸì1Ë~wdáøÌÌÈÌÌòžøþ÷¿ÿéþW1 [c; [3+'À2tGx¿>ïÄðdXØšÝ@·÷Ž™míœÞ·ì¾¾ß'Â?ÇÈÉ`ûgéßÄ `ÿ1˜¤ ml ÿ³ÀÅ`’ùC&¹?Ä `’ÿ½ë(›[üÁ÷ Æz¯©ù‡ÞsµþC<ïEuþÐ{æŸòï·’ÉÐÚÞüÏ Ïû^£?qÖw:ýæ0ÿ 3¿k›üAöwZÿ•Îòî0ð/|×3ýƒïÛý_vp¾ÇÍþù„ß¿Ï?›ÞeÍÿÂw“,þÂw#,ÿÂ÷ù¬þƒïß=“•¡½ýŸ ¬ï YÚ™ü5Ó»‚µ£1ÈÂÞé/÷AmþìzïËÆù/ÛÞumÿDßç°ýe{G; Ùß6¿ï·ÿ3åû@ö† ' Ck Ó?~°½«ÚÿuƬï“ÚÿA¶DÿýÿþÂwÙ?¶±¾O2·û+ü®æø'üîÓû£÷·÷ïÎ8Z:þe5Ë»ÆOX߯†“áß.¼/¸üuWÞÛwû ßõÜÿÂ÷î=þ…ÿï§-*jçæÉð>ëûq¾)3€‹•õëÿ•hì mþõZ¾?ÿ˦ïÏ è4Fø±hgÌh™Ò\æ-Q0]M .jÖ«Xß3׉°n]<*ç@»^§õT‘Ž‰²½Oìú‚ïÞá¥2!yî盺ð{ßÅ`?Ý£_ë*]Á}Ïžô>ൽiþæ„\iv}¬,Q»p0ëbè8O™Z\ýì* Xïg—†ÞŒ@N.-ÉtëO!Au”l$ªh©e¡ì!k®h‰ñPkSÞ–!p-t?/í:œ1ž àr~¹ÿƒnÜA̦ºM‹¥úK褴0 µöm`¤F8`¦ìC¯À©ŽÊdc×dyàIèÏ\|œ”Up9§oíÐ"y¥àÔ]/Pö+8·äUšn¸i‹·¥C‚ôP/Çìf]GôI¥uÄŽ}iʽì}ô4¾‡¾PöÅ;¶ ûôÕ£zT‹L«%Rmð—E–јÇ‘r¹¬íkš«Ñ4¸ö-‰­O/ãÌ+ÁÚv2A’tg²%åPÂÞ•'Væ^+Öˆ¹Ïß8KË]± ŠC¼mhŒ¶·³ùx„ÈG†_øȧ?›ÛiÂ+V¨í¸¹‘ôË3Q]ȵm¢”¼M£Eés=[Ïhæªë¥¨Ù_‡1•Êä9—®l¹#wãgv7Ctš† ‹6F³ÿáíé,í.&ö-Üo~ø;SÓ¤(‘Ê°r¯È(+‘·Ó&rùgLÔV+¿mˆ1uÛå²¥î 5™5n`㺳c€ ”8â'2%ä¾u‡Ç¬¹œÓfؼ?‚+é31eS+ÃU¢ÚX\¿‰Qš ·—Ÿ]ÙÀ’=:üÌtòðÀI¹ Ðkžºãæ=¢÷ ­L]¹ö¨€ Y¦AãXï{²_Nóê‡Ð¤à˜çVx{y‰˜£j×Ô¿,!¼i;ª­£¬ªmŒEEÈqÙHº"LÞ'InÇî“Mf2ªÂ¬\(MpR†@XÍûy[V©Á(¿Ð¢%§5[fU‚©8cžŽó¹À…~½ÆZ!v!KwÆì 93¹uh¶ûµ¨C7Ÿòéç¡ôàðrKWÀ%ÎF8[ô¡²æByêhÂè×ÂÖyRÃHŠ—þ«¶Ývi”àŒTEâ>ÿ/‘¿À³zKd‚ª¸Û[¯ÉÇö}åöÚ¹2œŠÓZ!7ý%‘¯AôB¥Üö|—m)µÖ¬NöfšJ8l¥±Ië) à˜¡†ò~J„X} „ªŽX[ŽØ× IE¿ºÖ8D7¿q pM…%¿Úø<»l™½:õ1þ6ÓôR媴<•{T¼|³4öðé _²¹Ï[q@ñ÷b™F„Mµ,CÔXᔪ ¨çšuz߯}ªÚ¬Û… Þˤ¶S71LÂj+àR˜R±î ?íNø-Ww,}sÙB°¢&ϧ>b~®EŸÅ;6â[ñr¦¿+dŸ”G| Žžv+ltôœakžšÐÐoìå²fݺòùë[$ðÑxá Q¡xŽº|• å'¨Z–¦·àænßh@äÜ,}5'ž½½M Ñ,èȥ˜۽aBüîSº÷¤ËŸ‹W‰í™6±‚j1S ïŸÍ^¶êϱ”PKÆŸ)•ËLÐoëöçÃ0„Ñ 2Ó¾ÐË¿lßš#øÒ0óx õv‡E:‹¿6lwZ *v;¶ŒêY 1â/7‡ñZëy5jAÒ"ƒw¼BZHL 9Ï>~H›™qþűãŒFBë‚蜥r·OÓþüÙ4DÜ´RˆSÜ|r;„þ3´ñYQœ³ÁÞÃçÉ*~F ½6&Ô_άurl‰€ueÀÌ[tÞ/¬Ê¥ã`ò÷¢Yþ–;e;žÃø­áE’·K¢ÄDϲö– •öÞ¦³¦ôsšyïþ*©ÈÞ^…¼@ o²™qlA×°îöXW˜vÞ ¥&oÑ9ØÉüî8`‘7œç—š®ävtäX›±+Á(ôžv•Rôó ¡’«/šÎâ×À\.ÙŠz”d[7Í/ÄôØtoDÁd|p¡¿æ¥lyØyZzë ÕF¥eDA哪O†šLtF`—,%Є~¨X³l èN p¤ìR @wŽ+Û(gZ;ÓðG²"vÉÞ²OîÙF´Rôö> KìÈ``dàž±³ð›jy+Jí¦o¿ ?§™¶ïÜ·EÀ'C­Xd:? óøC\¨|ç:¨¦ÿA§µ"©@¾ð(tísͳ—':y:i Ïá>‰‹gÃU*†üÄêS ¹£ƒDŸ­s5„ÞÃ÷¢@Åj8¤¤p3\ŸÂiçöømÐ]‰±Iøbáè“Ùǘeg‹dl]×!'st»æây3Ü…ƒós$ßW·oE•!糿ÑcCIð•pös`"~I>y.>ä´ÛºÅHró'¹8iífd­½ÜjÍsu°ÈÖ¾{øðȅ嫨À©ƒE#œC2#Ž*R?y@g×#£qÆy>@,Š¼p.Ìd~íé­°ëlaÅEÐ÷o¤YÖÕuŠò¯ÑÈ­Ç‘¢ ý1‹…ô<_‘É"_¾K¦dá„tŒ¹ÏTvÑå 5—ÄØœ8 úôx¿éã~ϦCœ—~4¨‹IÌÑÜ(¸lÐÛ¤ñMfÈ•ák«®äx'Þ£OŸeÌíA•$”MRNžüuG“Pɳ²¿¿,Eî»`iJÊéBtü¤…È`uÖdL¸÷,Š*Rþsò‡ïLÝ45±9ÜN,'V¦!}Š¾2°ƒ³t$–èOè Ðýú¶]|¡Ú9&a¾pM)¼ÝÉôëgSó–Hës½B9iŠ·-æzrH¹ DWª1²–’=Ó19É÷EF.$›µzºŽvàÄ\΃f÷¶º. wÈçXBïGìÌ•Öîµçp厮š‚#œÏÞ©xEmÉ _ñ=Rºòg=p²Íç*Ã4Ü7F¢‹€Ë§n¿™^/ú˜: f„Û²ºRQ5_½*0!úùîá„VÎiøâÙøͨ Ò)5@i`пò÷ª2þN.ÙšMa˜Zãù§X“ÖÊ%Ü“ÙYï?iÄQÏ-ðü–©_ÝOÄtдüGÓ{~|<ÚbzSC„™ÿ¼Ë,£~,ã…"²°·Ç@.?Î7ü«¡oƒBÿ›­ÿ¬)¤£ŽM ç ˜ñÚÍÒàa‚¦cLEå )üdÉœœüU5Y$ö4úF:¤® «MŠœ%ꀴ+Òaatå ¾¹¾lXç­[¾8ÁÏéÂ}÷…¡®¬¤#®ýú ½è ¹g¯i™o Ì€A±îO$#ÿ,a¨î«Œbxv  7¾ X`›9uEŠ .»Ì°: +yž6Âm52Çàá7Öz5ã1rҢͳÎÅÛyp ÑIŽx„½Õ âÏKßÀ?2&Ôš»;Ä \¶p*­ÞèvkÂÚU?ÜWÔÿht0ø­…Ç{ÍúƒKŒ‘ƒ4€àðÓC¸ÇÁåu×·Nxí¤ó±ç]5 Yé½ÀU/Wª©I±?qg+͇e‡»œÕ‹Š'«º^ã±GêÂA‡…×ô϶¨`Jjk¯¢Ö˜Âr"âòHÁ|äÒ¶—Ù5i±¨„ÈxjKÀQ4ªÛP¾*$ª»©ç5NööŽ*”^Á+¤'Gê žÎ¡‘°mûtx• ¨ñJ¸¡mµ\A¢†ùz¯;~šmC¼©1óz3âñ‹øþ”Sª¢?Ù45•›ù¤Æ‰èCÇ`Ô–·¿1؆Øï‡B&C¡L/m?&1ª4‘©ÿ“WJY—"hmÈUЕ5ÑÛ>õÅo#Íß²ºA^°W2¡¤Twp¿÷˜ã T6G [¥(1Z¿°PÜ·%h—knÁN@̵)ÓklhP6q~<—bÿñ¼òž|r÷:™éiè¦`V¼Ðm7ÏXCB Õ(•”mö£ís7Ì¦ï¨ E|s Síõ¼Þâ|J!ZÜ'òôBÆŽþb2óФHÜ ±ùœ'º—ZÔõQ}è’Ý•RK>¿^0šYRJ¾ÉÁ¥þ`¥ª§éâÍVh›þ£vû*·ê{«u7ñË.Óm0@*Ö½‚å,Ä,êmÞL Z~“òCûý/òØßý Ì­5…q‚Ô»î§c]¦Æºñ­¡eûuÖ¿h©樕Ü¥ý¾‘mU¢åÛyçîc&KY-½|’¸ŒºÀCïc-3'È.´9!^ž³zÃN~,¦Gn–GŽ#üæ1̶Ç@ÇتÄ«Œ;í.ë&“Q¬uoEhò'_±Ž:µ|dúšÈð<îe$Ako^p)KÁ¿&Y1?¶%ùf%í™ðX ©~8Ìp™!h_‘,ÿ ®þL›W¶%p`4Æ戯ZïƒöR§Sõ”‡‘.¼k°/få¤Þbh®”¿{!S Sñ{ú ¨{ÐÝ1Â}Ça?«è„ã÷†9i‘¶hÝ…án]IfË‘ÝìÐe©(TMßÆ£ þýùòèöØÍã¥o%· ’õÅhŽŠ_DÔùoýN€ pRUª;)ø?³¢ºËv$Ò¿RCaöK x'ø•Ýᦵíã¢Æòƒ¶æd·ÔwŽgì;…X"ø,[Ö îÊ·ÚTØù~ÁÉXP8œ|M¦¬5öY¨K›åp›¦h‹  3…|lL&ýi³K ŠÎÈØ¿Ó¿„–ïÚ2xShxzº¡­;òƒäæ×xá¯ÛîóïQÏ0ÇøP¯$ìJ÷ÚzŠˆô(…á?q<ÀÛÔŽ1ÕZu¯Ì——(ºº§Ùg©ÈÆ£¦½4©ÆË8úûtš)]0q‹^ÖY†à§À5ÎVï"ÍEº·ÕrìÑFJÑö¼i}‰ªhæV†ÑŠrÆË„¾Î”B€ñÑŒ"ß~¢Šéæ ªÌ Bƒ¡‡è®•_0ͺ•G–Y¼gsƒ‘WÆz~¾¯šaˆ§i%D¸Nÿ½L-ÝHNË]ÊåBñ·²«šZK3&áÛblelbCÇÖæ~`N_p8×>2¡î[wZ¸ÛEÊ°¸Ò´‚z ¼cì‘ÀÎAÌVóÐ1›¤P³$7ØFªÔÄCѺÈbIµ6ƒÞ‹BlÉÌd>Œ6ÊRIÖÂúFA[‰ü4 þvÞ—5~Á’ð "Ú@Mcr9Š÷[»hE¹\ƒ,cb'ý²Ni­—É>h¯‚µ2MϸL§KQZ‘¹DÙ‚=–òI”yˆ~åÅ„…½ÃÁi(R(qn¢›o@ùfÚ¢,t®áZÙ¸°šö¶¶# £ì¡ŸÌOÁQò;J^n¤['A½ä¸ö<êWÖN!Eúßœ/7Ä™¹ì]ç².Ånm™vdÖЋc{J9I„6˜[F‹G«vç÷)GÏj踈~SŸÛå¬#«Ui…JÝbºp4h›8Ë8nÄÁòÌ|?4‡ÎñZÀûÑý[Ió¹ÊØnmÚzóéâ¨Vé‰üCmsÔHlÚ+¹æ¹ý„„½È6þáU%ªAŽN•ÙfH¼nã¬H*e¾½yÛslax„I±ñÄ-´ÇCì›Üjÿ8DoóE°á¸î”õBæJ€Éx^c|9ÅËò§Üä™Ü_ì…™™d»8Èó/NlŠMUö0¢'²K×еÑìÎ"å\ªâŽã%̧á”Ö-á®Â ^(dE°\0ýÊqáàè0­Þ=Î"*¹Ài©!’+”K[?be¨¢Ýýk?¦xÀH€XµI­¬¤À>ÕÖÒÄ&´z­SeJ÷§ò$h£:Ò]×{×–®¼(éñf24Zšyiâ8iCWbç7‰Ä†ò5g×6z\®ažg0äO“ %g0¼5G³Tw(Ê£…»£ìœÄ)j‡FïæXx('fHî¡ÃôÓÔ¥|é`T«\ÔYÉÔà¸ÎÃFPìýÑNÔøÃá!’' )IHºq§à Ô“¤‘§ÔÇô|W·õ¤ºD¦”näw6¾L(…NižŒÍŒˆ±©5¦Í? 忾EÑÑv¨¬"v#×0á™j‹¡Ó ^6V¨IHÐL¨—°@>‚.˜05³¼éj^ 5²Ü餴x ƒ·§«ÒµäÍ.l¾\=’ÑeŠÅ£Œ„øˆƒ·’-8–ÓE:íãÌ]hÓÜ…Pìè»ÜÀ(0(Á%û9ïÕW˜ªËfM³ÈÏ›¢ÜZ~gŒS/j@b”´²¯ùMÎ;^þõ¶!GÉßòÚ‰LÙlƒ³H(ݳaëĆ:ˆl#IÀçï.ÒÅk1E¢uvG‡Gá&û3@ghì*Ïñ€¿]-y¬v·d¥Þ¾VÏ[í ¼"M@oÑbþð‘±mÄ«dÕWª¼»¶5)`>A(ç›ôŒ(ûê1L6zÖõÉÇT)«æIê˜ß?'¦bœ ˆÚ™-«@m®ê) v~•Ò^>Ë`ɇ££‹1¸oË¥ûø`*®9iLn<³¡3ùúÌÊJyl7l1-¾mB9·tn¿Ø×v±¤x™k™¯EG“ß ÛÒM³³¥mžüê’‡6ãÈJ™%Û2Ö´,¤È´å±¸'ÞØ|Z˜@kéÂD‚¢6“ë'\;VúZ ²0¯bùîï÷$¢×Pê¥õ²±O°W£|¶þˆÆ‡˜Ÿg,Y¸b$­>ôVäk¤ô8-6Ùˆª¶7Eç#?Q©áæ^—GÈo¸k$5§˜¤G­ùcd¦C»x]\}{\jMw1ÞƪhŠbnèÆ}é­]Ã¥ ’sÓ‹þp­‚uŸrs¶ÿŽ½þ«=0Ãé8Àly&fU'¼Hhl?èjrLÒQú‚aÓ%í‰àŒ¸¹¦ï 8)[ëêpœ‰ØpÑÓÑ+‹½¡Ùã‘š¾¶Õûl…W…B/Ù{èÑoP§8Õ 2ظ3S­E&+\+A>°E® d»0rÏKñ£+76×êíd¯wN…ßC%Lñï‚湉h¾R(ú‰eËçdÉ}œ—Ο­äâMg»É}¢ËRË+IΙì}ÝIs< w$ ž Ì5l‚A“˜ø_âù­¥Áj¨BÂh”ôëÙß6·ÏÙû!‡‚„raHŠÇüt4ì@1Œ=–-қĀ ¾zî\öØ+ç2þ®KŸ ÂÙâSb cŸÎ‰¡:Ml(xùœe¿ ‡uKŽË.`­rÝ×=BéZº1mÄ*MðM;æŠÃqäÕ"ú¥z/“¶µ½}ýïΪ™áºÐEþ\½%g'¶Ú î~ähŠ( ÏÍ9Fÿ—¨SuJÿˆÖ¯yÕH}×FÈqÎ2õè©”—7¡òX™ƒó¾BÌ%*Åflát3cìçôJÖÔG•ˆ’²4*+ªã?ØØ WÐ ÜJcÌEe\§öA°®;‹«w°yeN-]û–’ç!%6älrÌJ®Av.^ž ¡Ò*ãÎp7¨g‚”]Ry ª …šŠU"YYðÜüh¥4sÎ?Sö ì`pÛŸ½rú~”I:U‡/^~4›N¦½ ] q-(ϯíb­…€ÅBÜ¥tûÁ:czôªPS<Øõe}³›˜¢–×0ÎTóGØuƸ6ÉùÊŸ4yÒë‹?—ÇâœÕ+gÉçºÛ┲7iÚÃÃS|>9)2[ÃèL+ˆæ×6ä6¼µ1÷ÚUqüìHl­š”ÛMÝسü™.Ò™‡@iG/voæ°V ¸‰«)U»“þa­ÜÔ$…®ñë§Õ‹hY×Y\Ö½(&ÖO6EjH:¡ìvaϵ@W2¶§é¡~bêÁ_5B¤FAŽ¬@ðdî~4Ô\ØÎ"Ö|€Œó÷yîr´F¬ àŽÎ+*YºˆN¯66(NLñj2N­;†™]€;›õÀ Ó·éÎp— Ze&×F}S~•Æ?ïÂe/o.(¿(v H§ÇÒ¹Ve3Վ݇ÛjÜ/û•ŽNýp*Ù¯ãþ“ ›`püÿ›s°!b\h*‚®ðª¥3JrúE£‚J\ Zê:ºMg?*¥ë¤‰J©‚Ex98ñîFä´·‡@E²ÃÛûõe|zÄõkÕ4„f‹™Ú(h<¯Ä@ È8䌞PW$\³˜§¼ŽÀ bšŠ$‚Z‹› 7bœ%­ô æÀŠÆÂi„±@X\—ÚGéz‘X¡¤­rÿR—¾½½)¿"L›øÚËýû_ò.ýž&¯‚-[Õ”ó/5PV@qgíxÉÆ<07Z~d¸«§Å½æ™*÷–®>S˜¦×Ú˜ðŠຸ¦;Ç´z¢·OÞl:n‰†E+Aº†‘¦Ò þç¾>HÝqi5ôž»2Û¨"éˆ/†°ÒŠi0÷ +aeÆ“$q¬B¢nç‚äD#â•éÉTè½Sc©’¿mÝ%ÀÞNèÆ — ^&NCº£ÜT .ù9ŸP]×Âözã{ëÊWeÖVŽ »¾lÂÕ–g'HD*éäFÞ”]_/ ­J)<Æ’”õ¥/âæGidÈõãýÂõ‚1ºfü &À–§¶½w«Bs¹á÷“±¨c4!»ׄÞF½*÷† ¡O“­±Ôöµúqrmå°·ÐvðÍçÎÑz®»IÃÉ´®  ÎÜÁCì®óP:»…ÆC¾œJ^nm~}P¨íùB'01a¾–“ÀÛôó鲞ӆ'…`µYûõìcÔëé‚,¶Êó‡¹èÐÔó,Ç77 ešœŒô£Ï(6ï¤M–C†N*¯J+‘å9îòf{ŸãÄ4}¿úÎ|;ó錈mŸ1eÉ ¹=`¿¥“ˆvà|fw32驧"a`RßZ×B 6ªq’ø ÇqïÀå 0r楔‚7Ï&ê®I°z Ê'eÈè)‰õ.X:ø©ò … ‡‰=î13Ì­t><=Ú+ÝqQcàt¹ø) UUtœº¹#ÂÖP/«ø‚×3É_6´ŽÚÒ¼Œ$¤eOÅ8•?iÆ÷ –äz »ô‘<¬ÁWuNDª6s¡á|Z”1Ä(]ünHpð£Só¥Ûèé,‹E°¬­9Ì,9 RÏß„ÀüZ¤À×a…;ØȺG”_¶r‰§!màÅ<«m"€Xâ8‘IRíys¢NëMàÌè85š7Ç¢¬ë~za™K¬B?i=Á„Ò O#¨Qù- äõ¨±›r™q~?Aé¢Lr{%çÁ¤ØpIæ™ßI^þlû”Qþ(±t<§ 3BÇ0ÂÝY¸ZPérLñûª;ó²©àL4õ3P¢¡‰²ôuDlÞ+ Òèœqøk´4¸ú$B‘ r‡ÁÉ߈Ï*fù+%.ú9³ƒfW‚ ý4Ä-Ëö2 ‹jG3%Ú<€•ÞÖcÓ¿˜‰ðÚƾ¾odåø#r5Y,Ÿw¯4iX”®÷jl-®j@_g§ærÖ×8¦ª0…ÊWäW/‚ÛKf#½«ïw6ÎÏp^iÅI9Ŷ¿åµ(­D\Ž>ªãwaëIQoÑ›?/>Ö <ájgAÓrñÝé—8=o]nÜ<ZÎÆq" d¸~£÷Ò†ÁÜ]\ e½ñÉÜL窧0&š+ÙÕñ ï3Z2e½0^H¨šÇRà«Oq.Ô53>ÁÄXÖy´éŽ¸qK ™9gV[Ð(ˆ±YZ^Ô9‡ÃCeëÀª–kîD÷ %|¶s¨Ù%8nPÓëÁßùMœŒ£L“£Âñ´”m¢¯YßÛå^¿÷üÉšOp}¹ü´}Egþÿòµ”dÍlÞt£ä°6W‘OgkŽ .vÕ(ëÓO¾ñ¹([ölße/Ð+ ŒÛÙuÔfP=öÝ#L„£2‡dVÊ+¥yçq”ëÔ'V騯ü¹¨Çªö™‡©Û¯›nø˜Ò3…$¿üÏ8 ßD=¹ˆW?Ðs>Àsû>¨«S­ä€õîËx‡À=HðŸnY ¹`+i¿HÑ ù|6Æç’‘Ø Œ¶xŒì;ð’¹ MRʃ¸£8ŸbLqV:Û¸_JñRU9×Ø«.&ìê¥AÕw;¹›(mS=ïÚ°~ó`60ág¡ÏbúÝÁ…¿wDºÓóóœ'=À¶¾LÝ™¡ánoªiéCfƒ¿¸€7ÆZ6=yO’û&zÉã¨X½˜hÐYÅ0±_DÈ­ð‰ f0LŠUvvp¾ [[É)Ø+¹öÕK3´»ìUùÎN© ô˜†Œä¥æ·Â¬ õÙ›ƒN6 Ðúô“ö<Ð Ò[©bìNæ:C8ƒÇÛÏÁâÚƒó™Û }Ùe`\§]gÜâš^ÕßF‹îùÅó¡úwM—ñŽŸ^…µÚãÞ“]5¯æ0€Ã[ˆ­àéÖ”.*žëÁQAaY抓磳]Ù­Ø,ÃRoJmö K®YqÇùd=¾¡HÛÏ^ö==³ä®X9ɘةÓ]°RƒåN!W€B®˜M7Ö5Ž~p >Aw¿žØGÎ’v·Þ{zK±:r¿ÎYÉ@ Æ’¯ôC5TÂ|m5{ŽŠ,rhº‰ïe?&ê—;¡uÆôºŸ&¦˜lÚæó_Ñ@|Z¢Y%–ñCd2d"‰Qn {5ð¢ÓÈÆç–å{ÉÏäûa߈Bà2m¬§*o"¨¶÷ãBjFRW Οl³î¾"‚,H8­—"äÙ¡Ÿ #©›pë}xpŽbj?ݲ^ƒ§ÃÊu¶ðQ+sò“žÐÌV"¿ ?}²é0§dú^bx9š!to›ˆŠ…Jfm ðc‘]Áᳫ½?¬Mä¬ì_AkFÕ•UY -ïÁÙÈ„:F߯büùÔZ¤WóI¿M3ËjI_A¼/ðFmk±ÝÏ´‡ Ÿ&;j¨·/C0ža§¾uXÛVŸž¦(-ÐV@)p(=œT½5«‡§±žX£4~¥áýê¯òµýɉ“œ?Ï;(i´mØ[ûòü]>¢³\û^Á¿aî±—„ÀR¡_ÉC;Ct ùË`’¨ÆÃEµÁqΉÌAGæD>£GTk#ÄmTÌ’BËöPe†ù£þWÔ7¸diÙÑ–\"ãjA3ߥ"Cë¨èmÂMM¦ìÅkuŽ¡Úï,áß)%UÏ‹áÛÒD½Œû\³Ãj9d\ÄnöUMÚøò-…§.ž˜’bä1I§HýòÙÙæ=äUšo®„ñѤ¹ß‡Wj&ÄŸûùDÚg¤ > _zDÈ/øÒX¼)0Ö<"¬¥‚Aó„³ÀxWQêŽdi5‹ì®&úr¾—¢Û¡!‚û“Žpõou“M™ƒ ˉ»çj¾˜ª@Îðp¿HXaàBàw=%ã’™¦¹nS[FÛI/µ+rªÔ+®I»Lòö8lÛ9ªçUt7ì~ ¤žˆ¿J ²ž @DÂɃ;7Q•Á¼¡|âá\à"‘å3Ô\‰†¥L`§qÆÇšˆ<]¬ T>–pÓë"jn¯£¬a°e`–GHÌ]3Ç-RnÆüÙËâ§WÐzÖ«hT‚­17ŠÌ4dÈù ƒ~6¡pg¯õ_9ŒèÇ¥+Áý"5bù,âúåJJ*FÑ»¯H¾s節6yc…,‰#Qv`æ’î_.o´©Ê–ž¾ªî&V*xÎͽ Ý㬲|oX½®&¤ä¼]l¨Ÿá;œâ~™ •a±%rEP^åE¾>ÔŠXÜBfÒý3“†}[MtŠL|<õÿCÚ,3 endstream endobj 1352 0 obj << /Type /FontDescriptor /FontName /HKOQTE+EURM10 /Flags 4 /FontBBox [-32 -206 1060 722] /Ascent 691 /CapHeight 691 /Descent -189 /ItalicAngle 0 /StemV 78 /XHeight 459 /CharSet (/C/D/Gamma/I/K/L/Phi/V/W/X/Z/a/alpha/b/beta/c/d/delta/e/f/gamma/greater/h/i/j/k/kappa/lambda/lscript/m/mu/n/nu/omega/p/partialdiff/phi/pi/psi/q/r/rho/s/sigma/slash/t/tau/v/x/y/z) /FontFile 1351 0 R >> endobj 1353 0 obj << /Length1 1334 /Length2 8871 /Length3 0 /Length 9698 /Filter /FlateDecode >> stream xÚ­—UX\ݲ®!¸CHpiÜ­q îw§§q‡àÜ5xpw $8wÜáð¯uöJÎÚ·ç™7ó­ªñ•Œ1æÓMC¡¢Î"fî` –v°waá`åHiª)ò€¬¨44P0Èâ`/ r 8øù%7ÀÁ+ÀÉ#ÀÉJrµC/Îw0 0;C,íÁæSO€,j²·è-X_‚%=¡K+½Ã?Š¼1;0b²(‚\¬Àv/ Í@¶u3ØÅ“ fk Pûg…3@ ì †ºÍYQ98æ3€)ØbÊöOõröÞ›Í]ÿÇå†:¿t éˆðÒ¹ƒ½­çK¨lJ/¹À/•üÿ(ê¿Å¥]mm•@v/òÿšèÿrƒì ¶žÿ7àŸ þw€6øß)‚Í!®vÿí•sÙBÌÄì-mÁö› ÎÒ°¹ ÄÅÌ `²uÿ˶7ÿïÔ/ÓúWb6u9I1I5¦où¿|* ˆ½‹†§ãTÿ þsüá—™@!}vVvvŽ—À—çÞ ÿ+—”½™ƒ9ÄÞä怠P'êËáy!n€7boö€=^ fcµwpyYptuñ}9MPÔ6‘‡À&öéßÄ`ÿC¼6‰?Ä`“ü±Ød@vv ÿx9lrèEèýèe¥Šä¾8µþCü/Rzè%çÑ—SȲu´úcáÑ2ýãgYlþláKs¿ð%¯õ_ø’Êæð‹ÓÖÙ qtù+„Àf÷¾¨Û¹þUÜKrû?Þ—Øÿñr¾ ƒØòïf^Ö;þ™'×  .­9ÄâOÕœ/ªŽM øÒ„ãäüGÄù/÷K/_ƒ¿v‚çehζ g«¿2¿4÷§-àË ]@5ò[]þ.õÅäö/’žáK ^ÿÂÿ}îÅÅ<¼Ù,@î—cäbðr³ûþ?af®P(ØÞå_ß‘—»ó?ly¹i`°Ø uþ§ƒ™àGëä†à?©¼ñ¯Œ°â–1Jµ]SíhAs±°¶…C NŒ+5:w¥i¸˜;;äîÄÎám>ª#Ò'N1)3O;n&;i^ÍÄ:giŠžÛŽ”×AX­ Ӈܰʓ+Ã% ºù}™¿ûrUè%5v‘–(`ºõÝêºÓ?òðêH§Ùj†|ª¡å¤PÃNq* å YvÇNˆƒ_ ò³AnbZeé)Az@žR^{e|ÇÃõ×In’4Ôäñ9öƒ»?þ¸w‡v‚^nª½žDסåí’Z—ä®#b¯hëèë`ÖÅõ à; =6iþ£ÀSV)ØÞ·BÐzÝ`æ™ÐuZórug~žkZ±P?í¥ò xƒ¨L­›\1‰YO|' ÷ac32¢k^LxC²U‘26ü¤B20ö¦Yµ‹^Ȭ¹nÀxå+Oþ>õÇâŽSÏO«7LÕZqDeM•LÇŽˆ”Âìç÷Ί+6D´¢“£Ókæ úÑ̽ô¥ÖNy£<>×ë¬ðá„Hÿªé*æ² or¦(ö,«’†Ûú6¡6+'Í—6øYUÔÖᶪ¿ü*öQÓ³ù¹e»ÈÇ토½«ÄP ›‡Ë[ ¼‘U6¯ÀƒïÑ®nÁ²¬SݶñeŽåXK}š>¼e6âŠÚ«Q¾„ž¼³CY´[þMìôDý%üQ¦Ý=®Ìkj±ÔoD"»Z^–Å #p’ÅãïGb ‚ ©/ fî$©A3ˆ|ÂéV&énëÎÛ[cÈ5êi¤®è×nªèNËòjÊ0 Ç ±.K_*͘,{hû¼¼k|*ÁÍ%F/É%¿¨8"Yl7HÎàužÝÕÙ·Ùرï„!ë ‰ð ZÊ£ŽUªý/{O`i"þÚ<+“@XqÇ…s·CèÔ¼sé)Ù¸6³Fm-lÆB\Y¬VÞCz¹WJQ¼3íó›:á0Ð%š7éÊ;†±[¥ƒrµê¶öÃ;ÇŸ¦î!¦J¤ëÝI’&Ôiôš8î~Ü~c|!ŇçW7ô'©¡îËöæºsõ.Ï’YkªÁýwDg.ÀÔ rÍÙî@ô*¿$DV!¿Øac­#ßµéˆ<Êâ ‰kÁ¢!_²ÅðÚÉI>‚ ÓY‰  Ô¶¡¸SסâÞ:¹ˆøŸVYT\¿­<ª)Ê vŠÂ 二]¶R£7E*Eî+æn¿ë¤:Ë.zrr©Ú-ç:¨^é£C¨–ú Ãçý^{c#qIS_â!Q–TõHÑŸKMŽ³NmÛ·dž<¤D‚`I¡–Uœ`6ºF<bæ¼ýÊÎÄg.bŽ|•T‘à­F­°.&®°¬S»ØWÛ¼z‹£Åõ Ç©ÐÍÏhÜr¯øÒˆ­h2 ½‰­»iIU ’ígÆHkržÀ_Z8SÙÑyƒøó–p²6Y쀗uó°AA#p¼¢|÷UÄQ1YB)/‡3¡¬kŸéV*š"¼È°¥ŠÙ<æX¬ýöZ¢ÃÕM¡òéÖô©5ÆZðíŒïnלÄÐ>•‰yYq‡bÞÖ踓§ M¦ÀÄNøè÷±Ò&N»=¡ÚÿøN|ÿËwQÏLÌê_vS½§…{.­?*ªª™Æ¯•?­¥Çãœûdmw\ÞF²l6ÍbC&š®Æ¥„ ¥…8gç ]¹°é|»´*$*èplV-| >™ÌþSY±¤¾'7b¯æ“Vn)äÕû‚™ 'I,4é:Ÿ²ÈJ5bøídmQ­\ýæ”9–jý¬ÓÓEn»e²û­ÑW[±„à†$±=‚ùÉçÏvÒ[ž/æÖSZðL~5Ø Å¥Ó9‰ƒŒZåÈÎW|ô7ÁbÏOƒ$WȾõîÚ7„ÒPë¹çÉ^Äl‘³I° ¯jV½ÈÔGä!EìÓ :¨|—cÙ‡à‡&™Š´Ï™òI¯\UôÂð%¼ŠÞ÷n[½óLÙƒÃþ¨ÉÅ 3&vO•KJƒÞ7çÖ'Ñaœ¢ož0ã¼2@Ñ0r"HIÔUÉ{ŒTõ»WyöèN„ÅÆ:{‡KŸÙ{ˆ¾*YJã”B«·Q¹ÙYø`ϹÑBº¦ýý±9ÆÁÁ""™|ufÞà•éÛîå§æïÒSÿªH`7KQ[ˆ|uÇÛdœR/ùŠeÍW¹eª÷, &ÑPð{Ó£®¹ªÛ>ˆùF|Ž#O.¢ê²€‰U0¯Î ¯„[fAʉaÉÆrÍ€©Ç”ŠŠÒ!Ž£Åù½ÆÉ ÇþÌzÚꈕÙÔûV ®›ÈG{Œp”¯M~-¿™Í1ïinc'æ&wˆ§%w'9ñQÚ2O>m’hÍZ׫F^|on—±Ý¨å¨;ñó‡w©“–•å%¿È´-s›Íe ªLQ¢gþYA{8!:»›ŽñëÇC8¶*.3âS`.|Z„0&ý%•¢Ö,>/H7]†Õ>YÒ2Ïfìuâig2K2îÔ%Ð">™1â»P°ßM¦Ên‚W—°@K7£·õúøÃr7­%ËÕ{g0ÉŽÂÂU÷i·ÎV1SAäž@‚/Ò ë5RŠº·ÓžëMýZ ͶœùÉ‘”o}\—¥‰xySÿS-2¦B u‡É—nâš(üø½€ ›OÔëÙ`d XwÝB«_ú­é?«ð. ]Ô?!g¹)¬ð†è_Dþ¬ÂM예ìáÅUò“&ƒÛhój,Uó{ýdÛþå‰Q¶nŒš„»Áu¿õQ‘æ¡¢ºØþNÿƒé­$ºêTÙ]É[\æۘݶðÔ»H‹%©Ð]Aï–ñDÉBF¸éIÜãÍg²ƒ™‘è7]ÌQ½rÑé<°ˆ–,žì »ú÷ïÊ8< üïQOaÁ³Oó±õ7t%…¿÷ÕIÈ‚ðšq-E/¡ŽWçžtîû‡‘rO *lª¤·¨fÚÒoâà ›k­ ùæ*¹÷ 3šRûæz½ã¬ dD\éþA œ\š€'£„mWà9ηÿ9;Ï>ÀûŠfç“‚/Ì÷6å™ß;¶µÃÂìŸd•§m~³ÓKÒeÁ^' Ê]CXYá§-ÝCM¸UÙ艕3K‰ !›°„PKs5‰ÿ Ê”‚ñ¹ªçÉšEÇlë0¸µ—O+l§á¾¥mï«ó^V {cß³˜ðÆØñ}[’K0»CWp¥m´JWwæ}üK?×8žÄÔ¸;|öv¦U¬ŠYG@|WØUšo#ž[î"O«>L†n/Ì }öIÐTÛ¡㹌Êì[dMÃÖ½g”É÷¸É©¨üV‰qHÌ® ¥gÚL/`¸ñÕo¢×ðÈF}R a:÷8¨Ç£>Z† RE}¨g˜J\Ñ>~ƒœGÜTWx~scp¿óu8e3öáú«±«ûŠf¶×û{[¯ÜIŘIBPmBKË[ÃÂNi‚n˜Þ[½"¼.rÙeô+Ô&XÿçöožhO‰ø¢ñ“)£ÀÒ×?ï­¾+‚¥‡È±ò•£›dzî•~# ?rÝ̱Ó´ à G¢<–¨Î¯8­¿HôÆ”·EtMªèÛ\¡&'ðçÍÓ™fÞT³ÚyÔ{gV˜ù9TW«6JPˆ±Éÿ6–õ“Å:Ý©±ìÒVýBŸW%Mîguóæ©:á·ûšGaë)ì†&M Þ*°<²hGµIÏ‚ÃûÈYØÔЭft?ñ·Ñ|x:•n½Bl…j…ΰOš“{(BÈ➦8â/V ºãjò"Ù7 KÒçÞšØÚa À«t#Íl'çsK?àü*š®âÊ{ÙÄöQr?2VuÉv7Ãö1ø 2e„ä«‚–¤·e½j÷ƒ Fökt:ò[4–·÷Þ°²¦E̾ò×ðÁô…6\ÂgIØk·Êf·±Ý"VøߌÚ$©¾K§àÃ*B³ïE”.'h-º÷Ͼ}Õ¢µVŸDCßØhüô‰ Þ"´ø:¨WÍŒ±"«+ßL=¸ÁÕò-öÀû'áD·ŽðcMÍ×ð°ßË­ Aâ•°üÔ³g^»¤×Lu„ÎaX­[vHÀD.¿r¨(n!žW¬Ž’®½wÁ¢–Ϩ hÕj÷gÅ#àUv˼\ÝFkŒìÖæñdÓg^“ÕkÕ±c­ÓÂ6'Ž’ØG«€7ÉÖlô°&1“bV¾4+דñ—ª—Ø”Knq%Q1Äu¸µr G&)©± F„´š~ý1ê¹Ê½8Ü>¥ˆ;U Ï™CëUOý"=±Ç«òk}ò+RÝﾟ.¶0ŒÏyx¶¸¹ÆwªŸÓwN ÛÀw{m °±ôl<[¶*8;’i“Ì Ÿ²œ6ËE2QUh# ŸéU×2!kD³PKýb([Îi„ìPVÖ$µÀ•ËR$í}ý>]Ÿ[_2Yí“T>èWei€Úd¹løŠ+éñÚ5I|Î{Çæ0xrëUâ„Ž0:© û½¦æùÝû‘²®{ÃÆGi‚bè6i¤ÝD?öêI²x¶VGoûDQGéTßp|1ôÓ¥ÆÖ¹!_ùñSN„¹ *·¾Îà*¤Qv÷ÅÖßʪ¹S*àÝ{öz—ž™Q3Â%ÖKy ð–à×ì Ø\Q»Õj>Ÿü£ü54¢^J”´†øçþMddª/®>þØ—°”Ž[ÊñQŒKQ—fVgŒt 2 "Y=ŠÂó_WRpXSäŽØyßw;KFqc¸³EcÊÞFfÒ,_±ÚØg0ô`´*K´ KÊ# ® ÃU 3…½æþx ‰N˜áO(šãÝ‘Dù€{Ô`“¾êOøø±–’µ"tè™% DóYãtÂE ö˜ÌÔØc8¹̯{©1ã³ÆòÙúÍÏo®WÜ] ¢m‹Ÿñœ€ï6éâˆêèWÏà°ÖÙÝß'‡xÐHí-]=(–¾y°ó ‚ES¸ ¿¬ØÂ$çnâÆr†CÈõ™¢œ†²_‘¶ƒ ‰.-ñhíû"üh³<ðQå N+9Â’ñ¼íq@ˆŒåè™Ì¾ÑÒ' í.È$ ö¨ý]ìc²)+Œ‰&g¼)µaûóæª`—†`2Çj[ DÅ%6>ÀŽT ÇÊC¶5‘}¹¯GiH—véZ +¿DôU;Ò´”í«¸w .Уw£¹ª¿BòšŽJ°³6êuóòoDåg9 ¾ž®dv\i×ë(ëøÈ`4È»jY‘ hoF&™Œtüž¬ÔûQÔ2 JË":ìâ†ÛÅž4”ñÛˆÝ=ï Ê%0[g–‰Âp†õƒß«H‹†Ýè¿i ݧ7¬qBVÐ*Ú<(!W<·ÀGæR`JŒP–;¡Ö랈¸Å{p¢÷!(·ÖŽ¥]xí@œjÞµdºé°Ä:ÎL4¬.”‹™IùÍëë]9Άî`$ѱým—®dêFÙ¿'ºøéŽk08µ:p‘‡ˆ@¹³n•kªÈZeÿºµZU÷íLÍ› uΖ`ôù”®-ŠÎ_ag\è‡04UŒ$!ÔívÄ'¶tÄ}ç:AÉ,¹æ(Ø<=³v¯ ‘{—E™ÀÓ³Gò¦ã"iB]nŸ¢þ»ÖˆÔhk6lÇ =Í7M·—œ:À­GãZŸØòpÚµjÜÞ»_<ýÚZRÖö  ”¡…t}¡5“‰œð'g‡~©P¢~úím£?KÒ Jbt§OUû©¶º»óì“9¬âS•ÉY ñ?nH˜óÊŸÊ*ºíצ§ÍϘ”Ù QpoÍÞ‹G š”*[øÅ3K¼VâTñ ›WzT…HpÁ-|/áî ëÒ347Q>œA,f¸ª®[7U0ñ£ßYõÖn†ƒÉa'¸5¹ã½‡·Xl¨JÜÝܹñá"a¡~&FKs€oßF_9Msâ®êY·Åó=S¡ÉàÞ¼…ÿÄ¡þ‘l³úköûIå“û»DDÜí®…âæ9RŸãJDLE­Dìnÿ¥U'B‰.”w­S{™yôsФßJLëÏH©B|*#\f„sðQˆ¥Ø.úXŒúÊ•‰|‚Bg _}©/“¬¥³úÌië‰øòm†Æ²ûFš©Äeu5—ú©†zy ¿½‚0ñ`û9óv/tĪ¿/5 uã3¢q|†éõ+W•S~e»¼r¢å­_`±ùýÅ%Íõ³ê:…á­w~bu#z£*:•¸›áW‚„ÜX¬Ã1Á<Êé=vqHÞù9s4³Vqq²áXÉ^$#š¦HÛâzt³ö%ÛbÔrÃ̧p¬<úônLJµªÄªðöÙ4'“Ù¸ëåÍðXx%]x—ë«Ý°¥<£¤¦”É WöZ_ä²sì÷š¼»¼é‹­õt|+ÅgÇ')> «22ðÒŒÃTÁÀa_V(¼±!Rº=ÛÈ»p6x¦ZÚ¸dÞÙß7±ø¹+·æ¾ÑVnQHe­`RÀù´ù¶™ipvDúMÿx)>ÔHÕÛX`ݧe¡ó ÿè» Èø­u¡zÛI·¡—8¤YÒMý¶Ùxšë&‰¸–e¾-’—°v)ºÕÆûÆ)­.‘¥€Ó'¸¾éMß“ï•“íÙÒ)ÿ9Å{5Æ£­¢‹åÚ¶D«MÚ÷)_m[•xv†Ê—ˆbSóaˆ_è†îÞIP+ùÄÕZW³Ý°J¬IE æ#I$³FŠ±£±%æ‡Ô…õ³n7“c“ìÝwˆÔ~,øÜbbE¢- £ñB‹Ð½‘\OšÍŠ·Ë¼G¹a|ÌY|¸qlæ1”I}¶d$†ØÃÏ»Ž »qŸÙµïeˆrË#LºˆÆK0Dž¯™­"|+—ˆóøšÁœ¦·k<”Èø N‘h¬#H¤wÕÜi§ ªó0’Gm¨þø&>FÀîì¯Â×óas˜!ïê8Åίg¦Š~£,±Ý¥]FíŒÕˆ†Žã½­í_á±d<ìåYiTÝÔÿà!˜ ÷1'˜ydÞï’!¢â¯‡Þ¼o§k¹Þ¥„Ï–&ã’‘Ê{ëð3ÄÌ ÷:7ÔZ œˆZà"ÔR$ø±'Y€JøEó±âÖ+á“®œì‘žð˜'ý/Bo Ûä“¿fËVó¯×²S•å±²£¸8™ _ðÛ·Å*UO|Á¥Eœqè[׸ÑAuÇ*Ÿ5pëËIôÍJ^,.<Ÿf—CF›åDÀ×…ˆ><yy»³ë)Ø2ƒù€w¦|nØíCTe"0øÜ8~ð}ÈÕÌÒ 8OJxfƒrC@=ÜàNg˜ééCÀm‚´ñ ÑÒ Æ}·ëB³Â1C¸8Nûn¿ÄÊâóÑA=3KWÑ’åš:l¿º:p&0„/9RW´‰¨•¡L¢5Øço©Üäæ© P·÷tPЦ۠¢SE"åé†;˜Ùš€þ_Ö†¾¥H„ùÒ4%>BÈ+I¡—û'FKñ\¨¦N¯Ôâ|¬³?qèÌý0‹¼|´ˆPä&øºÄÈXè ”«ÅxÀ7@ÞmuÅÏ/#Oƒ&»ÏC™èö©LàñcÅl~RH¾ŽZ>ÌW›zx+¼>FI—¦GP÷glƒ~SXcÖøSµ½ŠmdÖÃ= AÂRaÿVwƒ«Õ ç8YJv‰V¹fôù’ ðQd˜ÃM]úLTì ¸`6ß Š+­O8&|`Xù‘éT\Ï’£Î61$J ¯â°ñ±2K×>M<ÏÜ[|ù–eýjë­±³Éà+ …ƒ-¬Y¡¦ˆdW•ž§¦rÞ´i™/®¬«ßì )ZüŠICÂèH—0ü'^ ¢™À+zZš¨`©¢lâÂ1Q¾IÇá'ô«Jû™ž„W`ïñ}ÚÎ@æwéûr•Çdj¤:tøßZ²‹ƒúçÄ”i¨ÝÉ´¾&Žº'x{=tÝw½µRMæku×ËË ˆ–œË•h Æ9è)CÖ.®þA¥‚×å_ð¸”"aëk0v3¶ :P@6õ¹XA/ÿÓTñüVQ‘ÜsEvòYèÁÚÈò±ëùUöû²4×­ú 8› »Ã%^=›Œ" 4ÃO²ôŸ×ãGŒ(&t‹T‚•«y|óèíLušn]¸V„kÏ %6<ê”C–„ñ²[dmL×1ªÈïø_¢–¹ÓºCž 4¯áµ}êusÊ‚E')NHËóÌå5ƒ[h³ø.3…fÞÌ>zg] úÇñ¦6á[†ø Þ¨Y¢°+|š‡ÓÅ hæµv¯üеóÙóHâÂø^tß“m~Ë5Ãh¢7~‘Ü¢Šç(ÂGoZWê#ÛöWY–~½—­ÓDp³»_KŸ—¦+±PAÔ|ÿüÓSkV4Ä+ßXçuEš¶²'‹Ã4ïÒsÏíD6Åm’ÉMí܃ŒK ¼©›æÒ;lE¥_d˜€õktq9±cüÈ['e-I¶°­à’^Yá5ñ”*ÜH) æ×g¶k½èŠßÛâI¾ï2„„†ºÞ›Àp"¦«7ž!TÁp$UOGÀ3êa â‚?»÷k†bÇpÞz<ÍÆÈaÑÏÝÃ"^ö s² Õ\`ò¨S5C™âq0¦Ä‚©ê¥&2ºÈ'.Ü噶;ÍÚC“ÁÐœÝÅ”UJN-Ú‚z'Ù¤–/2I>¸ òÞj¦f€ÛóöPlÜ:a?LÚJJ—k £…~8‹T¢dJÐÞÀNZùÕÃ$q«»æ6%äš½Z XJºy8B4:žRØY¯Ù/1DÿìNÍjlmtps£ñbÕÅgÜ% 8ýM´¼q°ÊsõÊÑ;ƒ*|‚féZeÂ6„þt«ªz”ÆÎÐ𥠸£Ž)Y6ÜdC¦nW—!ì{äâጥïaÜtj^$Ã'Ü%L0V÷³Ÿ¢Ú÷äj-öT†>LbæÀÑD¡Žå µ%…hPUïØÞ3 ÛíB eÜû·ÂшŒTZ’tBœ²»‹ÑMÄɧ•Ä tC3ï;d>Žn í½‹ùšËôÁAùKö—ŸÉ” WcX ’[X>ÍÞ±PÓïÌk. üÙlýÉ‘_e¬3$ÓŒ ÕyE£ ⸟Ç>£®$â1v~Pþe8ÊÌ'=¨×oåÅBxùµc"¹z ÐskîÚL9}¢g®!«±3º}÷SœðµKº>AcÉŠ…MƒCE`R†2¹IûÌžÐ2jqã:†qÄVA`7V‰jÂ#cÔDÜúùD?µÀ+owÍ̉ M?„£pÎ6ܾ‘OºMM]R7RV!¬Étƒ–pI4£ÔXVJn-%yÊ!磭ç6zD–¬Þ³¥¿ègGñ'Òå¢wÍÄÇ|¨a íùÚóééùȘ²­Tþ“ÜŠAîia7“S‘†D¿l4?€1A/ñ)ö®/ö‡ ‹lõ§Ûvº¤ ËNù8˜'Ì^ß*´=eÒ+þ¹PRvƒ‚`‰¢“cmÕ9›ã3¿ˆw8`Û׌$Îç‘tà™ qo™PÂ3¿‹ /ã»C‘„~qÊÝ e.}o[œÇ®wÀå(y~Ä#ÿdA2 £V,Ò>~$*V¡ÆÜÇ:»éîVÝ’!_RÒ ŸŸÏðuŠÍ.ûnx¢éphÙ‘ø"£¿Øø''."›-kN›h#†n=á´ÊÐÚç˜[ð¤ÒBPìG̻޾ê~¢cgidþÓø#"ƒi7ûÿ¿z· endstream endobj 1354 0 obj << /Type /FontDescriptor /FontName /SIDADR+EURM7 /Flags 4 /FontBBox [0 -250 1240 750] /Ascent 692 /CapHeight 692 /Descent -189 /ItalicAngle 0 /StemV 88 /XHeight 460 /CharSet (/A/B/C/D/Gamma/I/L/Phi/V/Z/a/alpha/b/d/f/i/j/k/lscript/m/mu/n/nu/omega/p/partialdiff/phi/pi/psi/sigma/slash/t/tau/theta/v/y/z) /FontFile 1353 0 R >> endobj 1355 0 obj << /Length1 878 /Length2 2859 /Length3 0 /Length 3466 /Filter /FlateDecode >> stream xÚ­Ry<”û¶“ÈV8‰^B$cfìK²ïƒ&k“e̼Ã0f˜»,‡Œ-KÖ”¥”­ƒD%ƒD$ QvrYKŠ,éNuι÷vÿ½Ÿ÷Ÿ÷y¾Ïïû}¾‹Œ¤íYE},É4!©Š0L 0¶?‹€A8Æ-#cHÑT<‰h„¦‚ZLSX“8€©k)«i©¨pËÆ à(@H,HÁ{A,à˜d_4‘œCûá ±!É/„Œ÷ô¢r†òß3ªú¾ Ašêú2 bÐà, ƒ©!@Ÿ@ß_P$HÉ  ƒX<† x€žx"·ÒwûæD PÿIcüþ‚d £@ŽÑ‘<ÀèK"BqÜJÖ$F-áäÿaê×ä&‚5Ú—‘þçHÿ'ŽöÅBþR|á¯Gð§%ˆÅøþ5§¢ xŒ>Ñ“Пžb‚±¶x*Æ À¡ 𱿖fŒëGa%c +cC¤Â_Kÿ´Eã‰T»¿Ò~WÿÀ°cÆTÈø`…@¡0†ñýýçòK1c"†„Å=¸ª€&“Ñ!ÜŒóa U à‰X0ƒŽ• D•ñð  ^`Ü™ûûÕ4%£ïÔ¤®(Yý1bÖÿ rúÁ  ©šüœAÑâ¨ÿA+ÿEÿ¼€ïüÿNËÀ€¦¨(Â4UM5e@]YãÂÉ0d2H¤þ8?ÆÄÿÆ8sú˜‡yë'µ³ÄG¿ÿêÓ²*³MÿäóòLç›íù:–®ÛÊÙÍsŽK2=FÖ=¾«¦îd’G°§]¼+«,‰äÏõ/W¡Mñg^f›è‰ð¦qÕ+Œ­‘·‹¹Š2öÎZè×C çÏÆìê¸48E8 Æd&ÖM:ë^ù–Ê¿°Åÿ½²´³Ä tàxÁ ©áÎqkõökЩÂ*ºû8šä I%Œ¦‹8yœJÆßa*c€nã¿Î5y@͉ðßo.Äé(`{±<Pñ™%ëé«ÈæbÀâS÷â5#›øÊ HyÂ[£U ‰ø©Œ1áUÕáSmƒþžÙg B¦“P)0úù5õ›ÓŒÕvv†%i#ñ@ëPê®û2‚÷bÂѱ]V:tùX¶Moo”ö"¢Cÿ xê.›]ZNòÒ«ýuQèìÚqŽ¬Lî@Þ†?vo¨áG„oßL©ØR«þ"ž@¿ßÕ’uËÓ‚9ÕšóÝdt‘E·<#Ù+Õ.:ÄmåSÙ®`¾’çm$'ëÃñni¦ßä_ʪ‡é MÄ«2(]·ú«ñ4}kv‘²ycÉL¨éà–>^]¿´×âŽùšsÍ¡“½ùOËÃb§ŽB¸i[ûº°£ícãÏ@¿¤Äy|Kàn**EuÃu6JS‘J“,Pª8“°(WǦOœe¼žT—’®]An)â!Ó8>Íæك¶ïðå±ýmªz%ºÏÝ?(¦ÙX±wõʾ áÕ;°Ëôž`þ0L”5&tÊ…Kæú|<Ž&ÒÔ\íñá€Z»ßòDØ_7^Xâ²é}Ž¡ÉÒ§#¼Iw Óê¯ËÍqtã+òô7è=4þŒG2ý¡ý¾_ÈB÷wŒ)Dn>ÁÁÝüèúé@» Ÿc¯êüºø?$ž>ØC&©=î¸c®WYÔûÆ(@îÃ#鵊Hï­[¶ ¿òñöž#;ˆŽö•:á)ƒÒ^?4™+[&5J >ž“œxC~Ì£Ù*s~l¢Ú²x¢Yˆ!ç© Kï*½³öLŸc"ga+~+d>>wÃÏÊ¿Òª§3C.eUë>}FÒò«ÿ¹ÂcoêXîFã"òÆÒ.­ý~zÐ%•´÷ŽDX~.Àµº os´yιz+˾ÿËÉ«6ªq •­vRÁE•+9ÍÙ9•Oo=¤ÛI龈¯üØ9ÇóÊèÏŽcÎ…û¬÷â=–„fO…5ÝL¨¼Í^ôU³d²ÄüÜöôÇqÈŠÕ2·Š*׳ŠæFž¥Û†ìªIìJ“WèDA÷LE‰  4¬’ÿ*³ËqÛä™|ÅÏ;‘.Ç’50宆zç »Ö”lz)¢qz‹õ›yl2•Péài+Á+ ¦<~&ò 2Àp€Þ|Ëâ0Á5÷vïd=»Py½ìëÒ[¹Â€ Ÿu9¡›ïsH£—˜-Âäªú‹Q÷ø(‰¡Þqu“ŸË›ÝøUóíV¡/ ÙÒ&œkÚL’ÒÐÆæv¿‰¿ûf›d6Ë9IZ㡱šM‡ZïÖý=¸¹-i*õaŸ·T9É8»ý¼—4Å°yHúncõ+¢~ñ`æ £ÍsÓÑZ…ܽº-{B¤Š§¹|ÆWYŽzOõPnS ƒdÎÑØp„"Ž”¹Yß·÷\Tª7|l.Éqè9,vÃBЭM¾f³Öž´,ýÚÝÇbÈ­@Ô¡3ýέ~ŠñðÃÀˆWäËryi¹o%TòéᯭeáÓmÑ>÷M(ËïtQ’-'Mä×~³H+Kýv³åâ#°J±‰¶Ýœ#ÒÃ8%'y}`_e%¢Füz½_Iþ™¼ÞR\mú`ä ãѲìI¯OgdEؾzMHfßÞŽ]ZrOP êî—ä®/Í“Ü÷òžÕWÄjôÛ>UÖÐuoä™g}OöS: üñéG“ÿtæü¢€Ç(Hò£3Ë T{Ž£ZÏÚŸ«šH|ïü úöHM_‡:êÄG¹¢ƒ®Ë“ kºô‡Î81?1w­¼Òv~þÚêÐhE¾!w-‚6Ì ˜úˆE¶Fý1ƒBÅÒeøJ;‰möìדLUÀÄu¿ËU[;Ìè(Q»Ï ÝÒ< ÊH+ÖÚY‘sÚ‰¼&]¤å¡R÷/fÛ³¬Z¥¾«sÎ3q”ê&<Óà·+˜ð–psˆ]”Iã8çÛü\ï‚e¶¦±ù³ÎbʱO«¼6¦üìßÄ賺 gwlYã|‡|ŒæzôŽúÜ]»idßí—jfjŸLXÕà*« ë·ó»™lnv@nÚ×Ù Qi¾«ð¦–5ö’|Ã…áJÑWÕÆÈ[s,UÚ¿®3sÍ®¤*àóñ"›:A'…òªç㞯ls +„(¤Rì'mËæ/ŽìŽo»^üR Ø“* Sàœ¯Ä\`=;n׃F3ÝáÅžn_Ž\æò.öÊøŒ$"÷+—œN‚ñßc=¹GÛyÛ5d<ûþµQâÁý"Pº- ù€D.G¨¸´rÆšf_éì Çä]5×o7Ô@å0«ã–™"o\{ ó1Ÿ£Vc7G+Ã…Ë Ä,ïÞÎ-ÁsÒŠ:^E¡Q™ƒäéºþ m^^@«à‚†ÏÂòÓº^íî7­Ü}¡   ÜÎÍŒò NòÄæ¦H¹§ºUÕ4ô-58ùÉXþJc=ß.î‘6ôÞzfÅ°R ‰¾ìÊpÛX+»ÅšMú6Õå÷F‹.ùª)ö|Þ®Ù¹ÌKOé×à;7Ù°î%fÁ9ò1ô}/ã„>ûäíÄçt%5F.²àêJ0j¹¢±JK;çó¨sãP&—™xÆI¬ürǦ#/<1-Ñ…ÖV*¦mÔ¡Ÿ^ý³©9 SZ‡©Çö¼+–WkŠšaÚÒT[,ÒÞKƒ×XlöN@é¤h¥BOmÐcNÑÊv¤ ÎîK—Ö§ï‹óˆ,jPò>5œ*·­-+®äfíó´³Ñ¢²5xcè°#*í“4}¨¸œý³ûÖ¹ëb]ýc­“ª\b©­¯}ÈôÞÏœåÝÔlÏâ§f³2‘Õ…¯)Ë‹@ω+ZŠÈlHõæ1 ‰™e¿(•ŸÆí·jòoÓ2s¤-b>R%Ÿ€«‹ž¨ìlåäVÛ¤>|‡[Z@KôNŒBËccíÖQ­>:û`ÊLËÓ2Å¥þÎ"áÉgèÔî…B]ÖmŒ‹o>æ±lëw<»‡àÈÔ°ßêßÊ|´·êmPàäÛ¡Ú3Åþw:©ÜìVÇ‘‘‹bJ:ußâÒ^âô9ê`Cn›sô‘ÁóuÔrÝÈûKÈEM½ _ËTæÝ—Ïñ”l……wÃÍž,ÃJjìb6ÍÏÍ•àoÎIXÎo™¯º4#½ÔkJ” 'Sr?šíÅ~bš‘©zÊ-6p¼ÊPÚL´FÎ=n_…¨iv𺠢:VÕƒŸ5ì×Ñò~TÊvèRó΢h81Á¶µ6…Üç꾘CÿsÑjõpŽ}¿[©¾üRQÐ[¡iÅÊ û0Õ+,—L÷"³Ž9ŽœAº´l%?²ú†ù–‚óhˆS«zê“i‘b<Ñ¡¶q=iƒî¤™]›"ä{L­êÚ»½œˆ ƒNïPGÙê­œ%+ΦJ@R0¾…¥'×,ÇŸv,;(ånW¦™Š£áƒˆøé¥Å endstream endobj 1356 0 obj << /Type /FontDescriptor /FontName /EJLECR+EUSM10 /Flags 4 /FontBBox [-8 -194 963 738] /Ascent 0 /CapHeight 702 /Descent 0 /ItalicAngle 0 /StemV 70 /XHeight 459 /CharSet (/D/L/N/X/bar/braceleft/braceright) /FontFile 1355 0 R >> endobj 1357 0 obj << /Length1 800 /Length2 1594 /Length3 0 /Length 2161 /Filter /FlateDecode >> stream xÚ­Ry<Ô{N7ƒi1TR~a憌ËÈD7ÛÔ`†²“˜å7c2f˜Åq]½a”ÜTÊM‹ìKEÝÈJ‘’B¶ÂEŠ;ôvßÏÛý÷ýüþù=ç<çœç<߃Øáà¨kNãP@<‡Í×Å 08ÀÚÙ‘h è£0PÂ’ ’ùLÛŠÌqÆÄD q‚}4€1Æ`q†X(°°@.@—4àÁ h É`ƒ4€ ¹þd6ð ÐQ²%' ”Ëdøò–ZËsˤ’Ù‘Ì÷ý%©dàÈ¡2A~( 0g±€CË<àȹA Å`“Ê( ƒÉ†ê-«'°éÀø[˜&øž ¹<ÉÀNÉFZ€d‡Í •h¤CõHÉ,P¢äÿ!êÇæx‹E"ûKÚ¯8ú¯4ÙŸÉ ýaÙÁ ®à7EDÆøÿ˜%ðÉ,&ÕœÍ`ú[ˆÉÃ3C@š“Oõèd\‰ƒlÚ£%n­ Ö#Z¬÷»è|{ò•œ™Éæ;…üÓu™¼‚1ÿÅO¸ÌÀB£1¢äûþçõÃ,k6•Cc²€¾ s¹äP¨äx$ÈÇL6 À‰`=›Ã—”~„䚸ÐåGÄîô¬–C+ÈX‚Hÿ  èQÈÜ•À¿7´°à„„£]}#É@Œ¤ÔØñ?4ª€ËÙü•‹‘¸ôÓ™OA0¤BŸwq¨{~;šQzâZ¤u^k´¶”£ìéνŽšµ1ÏNK±®<¶ Ôî+rûü‡¾áô;µà/[y ÕÇ>ÁODžÊì\|äóNV±Õ퓈: >óA¾ªôéô¨‘”}{_ãµ4÷Kõç'>üî°ÓÊé½ÌË«j=ƒŠk³Ã»áE,galÒ`Ç!Xfàµ8Cao0,-uMoKLäQ¡l¹Î‹œjÂç<ÙÜ3‹‡§‡Û÷y*ŸóÅÔ«Ò-Ò­+&gcZ’ÊÞDGŸÂA>lx6Hþ+ôô‚ ä‰ ê64û~óÑ€JÓ‚`š®Ÿ+^Ô­µy&qá»ê6Ý©ríO›Tº&-߀cÂZ3M$µu*4°¾Á_M/v·‰§în÷+#8\Âxð¨åg?>ÓˆO )— º=‡/½Ü4à9Z)5Ø&¤ìZsñ\¥k°U²°m-Óüöíà,Wÿ¬¾ a¢)»Qc|jð Ýv"{Ìüæhf›ì\FÿÁ^vž–npIñÔÛî¾>éÈKeþUusÛ)ŠÁ‡îÓÅ9Ÿ„_ájå¿;+§J=BY4*&7ºÖàéQ·2­¯MÝsïŒaSæ‚ÙlÇt8•pWJØQR3”UéJÚ“‹Í"Øè÷y}WÈÕ¤Û h²¯(á÷!vÊÅa¶®àÒ¹7»nÙhy6®Aœ½'zD˜®!Æá΀s¿vŠƒ_HOÚrž¼‚œ—OÕìQ¼&¿é`OÒœöFØ¢¡‡ÿª€˜·/Ó𑬱ӮÌT†OïV„ ¥g˜€†CÙí/œtâO60R—æ5hiFc¦ Š­³Òáê$ù^ÅŒÖϳÆpL¡1exªØû½yR³&TG mò0¬w½øJuUÃ]bÖ~Õ—qU*DÐ&å i£˜+´Z… t:·O›é]ܱ+'ì2üzZ•[Ž¸n:?¯ïÆåM›Uá;ÃùdÏóag*=¤W;áí9ÂçŸìšaëÓíø¯¹ëÚvZø˜ÄÊÒ/)a÷- «»)B9½Ñ“E–‰ûD£/. î¦6÷‹`òª³ì ¯ÒÝÕ÷Õ&ùK:V>]Ý-i.^7éîMêdåËÅÏFØ9§teÂâ#µfqâõ/4^ýºVã±ðôb]\ÞÖ¥m.gµ"m½ê.zˆ¤“ n-à%4÷æÚ„H±=‚î×JÍÜòý ñš&ì_³÷bÜ´­ÄÚZT2p$>²ô‘.¡JÚ_Tç.õlßÈÕ«o¹2½Dö‚ hž·&÷ú¼ÖÈÉ›3¢Ô"˜¥ÌÏÕ°ª6"WhYÌ£è³àܗΗ8&„8}ƒ¢¬QÃ?g®a]Urá“‚ç˜ ¼=•RŸÚ&nð]?®*ç#ï.F´›ÅÕï鉮×9â@‰‰PÞ-ç®Í¹sº?–_ù¥ãÁù«“a3uHÒÄHhª¢³z¹ŠßñnBâ}+Æù\;DÉhJ_½ýšÍ–®éC:="ÈHã°ŠËŽU¸êÛHèèº]®‹ã°ŠKfgÞJM:¤*¶|¬3ƒmk2ãóê·K'O­Þ$pÔíz7¥UýT¡ŸIͤ¸{ïäË(¬ÞÖa?ç5ñê»Ö44%8ïUh÷®ž«O÷ëvrÊwù«k“Þ¤¯L¤é’ùµï Ì'ì«'C&La(tþ*ÇÝ:d“ýjI¿ìOùeʗЩH.‡<(Gô·/¢ý-]´ ³‡³ø˜¼Ú¯v n¤7ZÉRÚ0²ñýáKðf³m_Iï¢.5D7Uý¸X&'À)äk…·f’2êK4V1]©ýXôËÂ3%­5G‘±§ü‘=J£ í„=Ûûà œ¨‘X¸Ö¦ÈÓJ˜¸"©²ÞãÑ-Ÿ ;ß[m¬+ºg?éžt/kæˆW‰’^¡æb‚úBŽsë¤ÌÒ)¯¿ö endstream endobj 1358 0 obj << /Type /FontDescriptor /FontName /MCIEGV+EUSM7 /Flags 4 /FontBBox [0 -250 1178 750] /Ascent 0 /CapHeight 703 /Descent 0 /ItalicAngle 0 /StemV 79 /XHeight 460 /CharSet (/D/N/bar) /FontFile 1357 0 R >> endobj 1359 0 obj << /Length1 836 /Length2 1227 /Length3 0 /Length 1808 /Filter /FlateDecode >> stream xÚ­R{8”yF9\#“µÙê—³˜1ƒqmvrèRFÎKèkï̼3cš±*L­l§Ù å‹5…$…Â:löKY}¢I$³$m)§jBûbÛïúìß?ßõþó{žç~îû¹Ÿ÷1ßì„!ѹTÈ›ËbðX<ƒHd<Øcñ(ss>DÂ\Ž'EÞÕÕA<€w8W"Ž@ÄáPæ€L²@$†ÊeÂ# >èfr : J€R ćiΊAxi6âÒ`H(Á"<\ž„3YB`åa=/æü?z‰Íó ~DÇ¢ðx@‡iB@…˜0e7ï̇ÃàçÅ4]Äû\ŠƒøÄ°BÌZÄ*ËaKÁ(;?.¢!“ü?†ZJî-b³ý(1ýâ¶ÿV§ÄÀlÉ"ÂKĆøKß@‹#‘!:,ŠYZõRØ0Äa²!€[LÁoX Ñýa!¶ZÈCúRid] ÂváÞä€ðP›?ïa¡èO9Â` ï/ÚyôBŒÿOŒl…‹A‹Ãá ò}~E-óâиt˜Ãö'@áó)¡²'À·xsè@bdb;,‡+DZO$<ˆ5ÿ]\€Ñöϧ2NöÀŽ9³ŸË‡bEˆQ6…#\¨ÿÝçöí\ñ·.ãèäðx€ÇáþŽ&âó!Žpáre}Ž0²ZC4TO—æv$:ûúw%‡¼ ;.jnQßάN÷«¼ñ AWú(C}¡uWì–þŠ0Ui®zXsxÓcÁ÷õ mÞcI±éò‡sÃqû†sãkÃÆsÉ’!žÉ;éË•u×;''¨ï¾ß§D®hÎ{sëe¿•gðsí'›Õš"⪚ÎqróÎe‡¤­°pب/-9æ˜Òw@_–¹¼¯]z(:E§Æ¦÷-·^´ZU¨“Ÿ5ùÎ0m­þËæ‰'Ìj‹¨€ÜSñ‰è#Þg¥Œê×/³ÍÜÿÕÉ3üôÂ;ïÙ»¬{ÍÄ–$ÛúN O¿ËEçÑͳY]ù¾–ÒkhEb¸òù×"ç;'}Ë{L›~#0Uà ]ÿV–f`rn«RIÆl,ÅN?U˘âÞNìœKÞæ–N¢}"î·åx^!TÁk,ûÒ8-jv`ÇΚ½d…»WeI)k™¦ÆŽñø„ªò×­}kß a‰y^¿DœËèí/¾¥ „-wbëjÿñHÊLp`g=6>U—[TQ£ÓªQ„5ž©Þ¦JkP<•­aLËÍ8²¬òy²~ï»Yß âîI­{º·®’l.¨¶ë%?¶/Ù•äµ:àÝú4ø.}0šm°íI§=³Á­Êx œw)|vö¬¬bècâŠ4a»”Aª»PCʸŠþ)¤o•³àç¸æ¡¯]Á¥QKìlïV¶™K¬&.®ñ<=/ßU§Ú1è£l0qËô3•MêWý´lÊòøè!ùQŠ*jøzÿe“ÿ êóÉÚj-ÁÜøD3¾àv„>N‘Ü*:¬oPîkDÒÓ­Ûó~"LtãmÈ%tÛ-¦+~‡ç÷gÿè…v endstream endobj 1360 0 obj << /Type /FontDescriptor /FontName /YFMQYV+MSAM10 /Flags 4 /FontBBox [8 -463 1331 1003] /Ascent 692 /CapHeight 550 /Descent 0 /ItalicAngle 0 /StemV 40 /XHeight 431 /CharSet (/check/greaterorequalslant) /FontFile 1359 0 R >> endobj 1361 0 obj << /Length1 843 /Length2 2026 /Length3 0 /Length 2603 /Filter /FlateDecode >> stream xÚ­Ry<Ô{&%FÑb÷³Ë63ÖȾe™A¶’’Y~Ã0f˜…{W¶¨ˆ²tD¶Ô$¥”%—d»„,)Ù•}wõNz{ïçíþñþó~~ÿ|Ÿsžó<çœß‘“rpR5Á’Р%‰HU…«Áõ¤“)ÔÕà9932ˆ¢âIDsÔ຺€èÀu8\ÓÓÐÈHÕpbø¡IÇÒ¡A ° ïE±šP½AÀÄ$ã1(â.ôcébPÀ‰„ÁƒT†KÇŒäÏ ã½¼©€¢Ù±of:ÿ£0!ÇoÀ¤€ä@«Ã,CРž~›Ìšˆ#:ßÃXšÿT H¦°†YÃX£bIDƒÕ8µ#±¼@V'ÿ¦~·¤v(?–ü÷mÿ#òÃß4Hþ™püÞÄâi~?g­©(cBô"€ì{O±ÄÓA¬žŠñp(܃DìÏÖ¬uíC-­M]í•ÿ}»IžHuføÿGö{ÃÿƬ­ñtÀ¦ƒÁYDÖ÷ãuþ'3 "†„Žu-mE&£KJ]K x"¤ Õ1TH¢²J5Œuddȷߨ­ @-¾…vÑquø©PÇ¿‘6uÝEÿœÖÔ”DQeYªjª³î^CSÐU‡…ýC#“A"u÷€X;ûqxÖ†Ab ýoI˜—|ÒŇ[Üî(Ù§ÄnêU•dWñâÏg|yë’¶ÎËL‚Klt¹¼†”#z@qœfìPêõ½CíQá>±\•H54ÍÛ\9);çV·Ì‘›SÎ|ì5¶gUÇZŠ&ÑE3¥Up Õ%ñRnÕRq•¾¦(Q¸ä/gÕc7©œØÒRipçü ÑÏU<žÈ…tHW«+Õ§ïX††_„q`<ç® m”Ñe§k$²^ÈÔašïy.) øåèCÓôËÓQøˆõìé ÛpçæMø†Ž§L lÙàc‘Üý*>s‹à3W|Oºç‡ ûg:ÿ1âZ1ž¢âªl;0^(åxž0ûÑÒÇ>Ôv¥¶§af•7 Ú¤±5úâS¿¡ýœnà›Jô /4òžŸ s ü…‘¢·¦Ü$jV,–[ÂÒ˜“¢+ôÓ›ÏcÖv Ûb#®Gè-Û=SIᨵwœêÓ9½¾È›áZ…[KòFZ:âÕepgqs°K­¿ÁS“Þn§ftE$kú¹ÖCñêɆ²ñ@íc±$ß|`eú ›eBBw–xÁ$ïÆ©¦ÕÝ>é‘nŠ˜ÚÅ¡œmåCÒ2bO€œýI·î›Ànlït[8êÙ ÜÉ%Õxwíþa‡mIå¶ö lU_GÞ¯äWtÔ›ÚI§éB8Ë°Wþ˜âŒG­7®f—k[]/p™žy3)²R+3•­]´ßeæLxË—uD±mòÕår?'àñrûæO[ ßsc¯ç€1#ê)ºTzl¨^e;²B1Û¸Ï~~eˆ+mU_FÓâ^HĆœg>ÜÂëð ´*Η>È=@‘,ÈUÆÝoŽë÷À"Åx]‡!V~*¹–ýË az‚¨^ä”üùn¥ýü9þüKŸ† ¾"Ý¥¸§#gL&åKá{ŽOét®ªf=z3&Ú³…3îA·ê<ÈX?®²¼xŠó¬TKžÛ)ÍV ð8—’Úg]\¯Õl¢d´˜˜ÙÞŽa$ÛʆAN.ëlw{ïØï XÈw(ìo•ø ß;· `¾ðE“{¸èËzwà~«xcÎ\ÜuƒŽÂÍÍ Ey‡)çØ3JÏ”Û+>¾?œÒÔ¦üNìdMõ,‚2µH/¾>¿Ú9ÆaUt+l-à‹—Í•œÖ¹C‚-(ˆ ­Ôõ©‹'V0Ë<\Tq% ”4±Wósoóeˆ®Õý±e:;¦Ì¥ËÏð¡#×ê_qç²_Û‰³(¼¶Úòîl3Ô°Üã}Øï祵¢3ÚŒëó–)Ìíèrîn¦WvwtD9G>ý­D€©…±Ý3Ãos$ËÈ6%ø¸,·¢°Õ¾¬ÝkÿÚÔÁã,O˜ICºV»`½¶j;“`ô1Æ®¡2‰™Ø)†Pnù˜S«[ű-ôºyk•]~hÆ%±2×[ŸívΩsEõòȤô£ïÍìdÛ?ÄÖ"ÖãîÐ(ÔžP³@nêCÇ:§›‘¢ o’/ù†Î“¯¾©–ÑõÕv¿£w-Ì!fôj˜a¨`áE› ) ìØþQÚ‰Öj"Þû=·ëƒ Ù%*£P›òl¹ÜõÉ!”á'sÉYô•n¦. ¶ˆo…fB,sÙ¼¿XÍÝKF,æÙW?xîñøÆÂ#6§¬¬dñ¡à&²L-$Ñ/Òa´ù)|&û|å`½³l¸àG#CEÞ·qO¢/»|’ï›CºðXçAûÐ7T>Ë¥mGí4Ó¸ø SUaÉ9!žžRñb÷òxJ(Ý<ñ#^Q£Ïä3*Ș}y¢M\Y v­ð2J¢ÛÛkL÷¤»Ž±F&‡;®ñ?¹0ì9!óÂÐU(I<óHÆÐ ¼J=óT«gÂÒ… ôÜ;ÉÒIw°ûîlI'êù³mŒšs/³µÀ»/ž|ß8ã…×à|luâ f-µv•·{¼w'†ÙåØsB":f¦7Ó+AeLoëê´Ø üÝØØzðû>Tv¼3%K’ìÒñåðÖž{B[åODV 6nØ7©<XjÐól£ºõ5gûuùÜ´k¶Ãimð2 B‘É£!kË šVÕ믖D¿úlÊžÛçF09 ¥m%rŸO¯Ò$œO__e³½ñœ9GØ—y~Þ1¦5–¡ówo‹l¸·å¸¸G5dö£¾ð6NœÄFd7"{ž›ìª*åH9O[P~hH4})ŠZ5GÞLš˜}*t'ù4gh¬–àñì—åÂ#S¤JÁ/® endstream endobj 1362 0 obj << /Type /FontDescriptor /FontName /FITBVO+MSBM10 /Flags 4 /FontBBox [-55 -420 2343 920] /Ascent 464 /CapHeight 689 /Descent 0 /ItalicAngle 0 /StemV 40 /XHeight 463 /CharSet (/E/P/R/V) /FontFile 1361 0 R >> endobj 1363 0 obj << /Length1 2417 /Length2 26418 /Length3 0 /Length 27442 /Filter /FlateDecode >> stream xÚ¬vP]¶%œàìâîîîî.wîîNpww·àÜ@pw >äïé7¯{úM=™ºU·¾³ÏþÖÙkË:)¡‚2­ ±¡‰˜­-#@hhgK+dgm ``ø°00À’’ ;˜,ìlE€N&\1 €,ÐÀÈ`äàbeäbä0rrp~8ªÉ:;Í>|˜98™9,¬,ì,°¤s€±…™…ÐÚÚ`bkdglb °™[Øš>À†Ö&G;S'W ƒ ÀÔÎ`ï`gìldakp27ùƒánob 42qX[™Ø:~8ÙÜ휎#;{w 3s§+…å_áþbPvwt2±q¤Z[”þø8~ )™8š8¸˜Ó}Äöñþ¿üñüqÜŸÃíMœÜv¦ÿüÞÁÞÎøç( ­ñ–…Óßc²sp¤ùcü æ°µsþa÷7&&Æ4gÇ?ÿÆŽöÖ@÷?6vƦO0f#k»?¼>Ø;9mMM>–®NævÎNfâfï`âèpu°pr2±í?Ð]€Öÿ(Ý>Ý˜Ö Æ@'àÿ—”3d,>X¼ÿ“úP‹¿òíèlèhalt°0ù‡Rüs!þŠå_áß ýèþ¯¤þåß’ÿ?NýØ¿fþ¼?Üþ¯?Ñ™Y|ôÔŸc>Ž46±:XýÕÿ‰‘ñ#h£?,Í,laéÿ̯¤­©ÝŸ™üËnìlÿ÷=ÇÙPüïy¦ü«ìl?ZÃØÄ–^ÎÎ飺 á¿ô?˜ hý]ÃÿOìþ9ÂÙÿLJ`ÿÔñ¿)$°ÿBHþ 2ûÿ–‘ÿ·ˆÀþçDä?ìcØÿ¢„üä¿•èØÿ¨9þeúaÿ³âñõìKµÿ!á°ÿÍþO1þçys¶¶–Ú|Ì,#Óß… 𷛉™éÿòÚXX»ÿÍÿ/çvP7ùÛlýøç=É°0´5û(7ÃßLŽbn&Æ NFæS µ£Éß쪶Æ&úAÁÎÑâÏ×€–‘áŸö>êddeû'¬Û2ùÈû?Åû¡eEK/¯%£¢¡Hýï>IþrýÓn)e§”ŒÿÍð׶ÐÂÖéOïý=à?€­ÿÏZèä`áÐføPDÆÇßߟtÿE»2äIprrlâÁtLÇ趥öÚ´ L€,˜‡LSÀ"y;pN5x‘ý¾Ù jÑgÄ,?›µ±kÖêXº·#+"·t^y>?|O# š ÖÕ˜â|hBÞ°Ùí\£|ö}cúͪŒžÉ,íÀÎj3]€Â º€ŽÞ<ì_dl4cæSªÏK8”˜ñåÜ™Ç/uÿTׂKªPRæy©‹Tl”›åz‡ ðà•çæøoWû\1Ù}èŽ ¯Á"ßÍgìÀ¾²úýŠž•dïG]!†½Œ’oþ¡P‹¸åÌèû¿Ý{ÌYLíÈ'¿Jîù«ð¢s8@€ÿ……ŸËÑnÑ9ÜnßKÙ£`[hɯh÷)Qéºg(%NQÂÄh^•Ýç ØŽn$ܱÒÄOV»ƒâœôºçª;}.áQ68±÷ÅbÒWeµúUã_RŸ ñJx³U~/* ͬ/‘–™¸Ü‹È‰'É®}ò3w1V«o¬j#¶QÃåº ­÷°ÜuV¸zw´ÚÎ f!¡Gù"ÏQIg~`zó”Em^~â5u5#ÐøýÃ툆©Úov{§Ÿ{»®L:„Ôi“bÕB„¥U-Uä[á‹ËLÙÖW¢l"'®áï÷ÑÓ1¡Æw Þ%ðøñÜ0`zs4ÖDˆcäq_\Ì"sœ– ¹û4÷/®¾g°½¿Ÿ„)—l L}iîw'ß4SH¾.¶ì|[I¨®ÜÒt[æ6o˜[ê»›Ã×Üm³nÛ³6ˆà7¿z_7{«Ñ¥þžq°Ø[Tr þ¤p½ ¾h‘¢ÜóŒ2+“¦Áˆ5!ãöŒb‚ ΩK¢³þ-U»åþº†Ã„ü./DÖÍëÉ°ýÈ‚#Šw²TB{–(§ôTÑoò cŸœ²X q‚'DKøÝ*G…72¨¡¦0õm~W¸‹E~šú¾µ?3̦¥öÂ-5¦ž…gJ> ªHy2Åv¼Ûº„ÛÔ¡âæ¥qû;dâ“ë@š92à@b<+‘×mÖ¶þÜ…È#IóE’žÁg?ƒ]•zǧ ³ ªÞegM\û×ÌG‹Åð;|Rº.ÐØdúœ}Ö¦´|j-Y½k›rš2ªóßñëB¶ÍµiÄÙ÷Ûb+F·Ð‚î[uó#å°¡Éœ†adR=”Ô¿¤©×©_ûÞ@cnõ‡• ÊÍëæB€däãÉš6LßÔ J2·êʺÔ*ÅPöF\ö]U$ñpðÃâ׶K{«_¦‚%_ÒÔØ Éô~—Dè¦ÚyX¾6ÏÜ#±¼æcÇ÷‰\kH¦Ý/Ã`¤”$í¿PSp‘ã2ëçó·7'?D,ÎR¿Z 2`éB|)ÚlÊ C4<€Š?LÍÞY¤m8¬x!Á éŸÇ`ÀÎ Ñr¢0ï©EL›]6íµÄ,"ó³ÏË#´œÇ¨GwÈs ¯1‰ ô^3\ÙìâÙè[Ë™ ؘ)"•ö‹²oaCökœ½g9ÓšCÂR³äqNïHì$Õ™¦››º—¶’ïQÚ*†˜'ΚARNw?'¡`˜"Wš5^Í ÚijٴSÕr7è ñ~úºž{³5ùpm:(,þ8†“y9‚—¦³µ¹™Þ3üj©ÕŠšŸ+ô‹–ÖC×õ,ªÖYé"KfÒ“áÖD¯7td°¥×7Ò5IÅ”ð£[Úç<µ9ÙZ:ߧÍàzaŽ†žYé PvXÊ|úS²^·Ô÷¾8ôAD¼ÄŽ(yŠ&½saóM-B÷]µ‘Ï?«©,Êvj¬l‰Ý–¤ýøœ¹™ŸýÕè”gë² ) f°0ǯ>¨Õ¥ˆ§ˆÈŠ{®¥N¥“ÈúPÝ mܾòÕŠ¹vMX‡.çp¯}Ýij$f4ÈÚ Ë˜Å MfKnÔè>€0x2V³>‰\caÛôÝÙfúŸ%­|×ÈÚG ƒ'hÍ7»²‚+lÿYózŠ-½¡Â™ qÀfUð#tE¶k0—(BÞiÚè*r;PÍÐ×¢¸Lºí.3…v àš{¾60(^ø)_ªæ™Ôæ‘ Øåø íq„ˆVGÀEùX¥9]BG³ŒM™Ûþµ…@r>R%a E&„CiF’7%¶*‚ú] g‰ ä Œ]OÌéqDšØš¨ Ï¢>DÚÙÇS‘ÌýgòÞõk¨ŸßH2( ß!×›ârûöeF‚}.sÓõ» °ÉK†-3€W‰ŽøWVJ…­S¸]>€Ü (<„fRo%0XrÍn[ö[¼°œiBCm£æ¬Áƒ-žéïÍe7VEˆfóÆq©gÇqLͱŸà_ÝöÍd5àxÇ8¢àÒcOx ë)î¹÷ewzöE ¹$9ˆ±Ó÷¿š#˜.lŽÚ~êäoXÃq«Á¨ÌÆ–H$o!ŽR^¢ü}<švF‡—üýó•Í\ogAª¢B¢àl0þN†RKHÝm ÚÐ÷“;á4Ä-†Wd&}V j¢>+çÍYN$p¾¾åu>gÙ§½±v³™öÈ%€ȽúàÛÞe‘e‰ÊÝÛèáQQ¥â0Å^¬çr¿CÚ¤’]Qä>¤%½%ÙŠßQ?îWRÆuH2TŽ)5+·‡7u§Poxì›ýJ_±£'»}dÇO–ôÃ+eëž‘žAäþÉ„(â1Rì %-+—:Ë ¡_–/xÕJT¸œx§ÔaßN|™˜±`{+Ê„¿bô@áöªXŒªWdÊ?)QÿÀ¼Vq°¤™ÔCŠh î75‹sÀº3õ?&ÙŠd:%Üøùê§Ì¢¾8ÞúbAíÀG)æ Âeú•Y—KËȱøNZŠ7¯£ 4¾«B¯àaC*°(¹zÊ \pGÂÈUÌêÛ¯;‚4šÉ¶äÔs«yOr€kæÐé2w †èÂCð5dà»´2Ä~ë’£ŽvõšêWFG߀õ+!Wë=~¨âÛ#?_Ï©ZžIé(-éÊ“:Á±Öö?»[mµ¬?þ‰kµdÿn'—ÛÃzèž>ó­xeôëà´wÁGyY}L™i<-¢˜ŒÊz=Õ¼gŽDÓ-û{gÇ@Qí½£È0I¨eúP²ñò¶[ðö’ì2ö¥xu„³Ü±Ñ l‘uZµñS`6eb?¯-Ÿ,w+Ö-œÁŸ[ wtÉèûù@»—˜Ë=Š©ãT­,XÇLÚü,fžüÕÂóûawžr|ò6‹( ŒÙûý17qT2ñZŒÄÙpMœç+½Ó’e‚µÌ«šÖ(±LÙmÆÓšrpY§#fáÞ1ú¼ý~Wº4HïW²D±Ïle·C؜եD†ºÒiÎ<·]‘tüm`—}à}Lä€Óq'à†À÷ƒ®óy OÝ{KΚ\ì.4:ˆ{©rcY¡=å‡x´„Ðܸ2kä)öZ1eh†h¸„Ed)yîJ-r¨v±ÃágËñ „‡µjõÝG·°QïÕ"ó‚Æ£áUÀìæ%kÅ$¢hÞݬ‹tiij¢ í5” ¶D:gJóYç@W8y^ ¶—…í ߯hGŽÄ7OºÖÉûüAüÃx<ñü9˜w•°ŸaZï£ú¹wF ‹+„ +Eo' ´ÜÝwí­ruΓ¾düJB“/3÷·˜„HÓèŸR;MÒyc@°ŠD¿x;î Úȼ&‡¸iu¸*_U~ßîh}hŒ”kã&JÎe¿ë¤+šãC¹ÊmÄhÛÈ}mÓ°J¶º¼y_¬óÀQ pÓŒÿÒ¶¹AÀ‡Â ªU]žÜÓ²UßJ>ôDâ\½9sÐî9©õgòàn’AÉ*úeV¨<¨§å1Ö´ìïKhI!¬³ño•kû\¹:ùÍ“`X¡öé¾Øî™ì5<™ùF»´uàâd÷ÈÍ»7¢å9£;¾Ç·á½ZQ5.®™â0¤ Ä¡°äIQ°É¦éúi³ .,˜} Ñ£dj{HÂ÷?ÒBô‘xlsÑ™ƒ»%ð–Ë ¼“Lðyõ£^Sq¤sNäÇ<òêDC +C³pÄd¹pÁàñí"I8˜Ò†SÐ`Ó’?Ô3NÛX†•)Ž—ðs]v˜…ŸöݦYt]~-QÌíMý:áƒßnõ]?ÏÄ·M"Õ)Ž¾^ë /3šÅD ïšÔš`çsÜ j¹æ4àgù\BÊ~ð2ÃÙc9°/5Šßëc)J1åؘÊÀu¯Ž¸`Ë™`‹„ÅK ˆ›=îAt¸éœÀHŸœ9ªu=CùÉÆ|ÕµŸ…&M LEHml4=ͽ¾Àˆ¶. h:³\œGÙ Öû0ùÍØw¢Nó¨áhŠºr.Å&%Ò Í•,ŸOGŠˆ™pÉbjZ¾Ó¯&ÚGY½9æËIr´Ö©›¤öø1Ìú‹möÓÃÏïìØ°„L)Ö*ïu„¢ä×5"ðÕaÔóIÜσäµò™Î‘ ¯úµ¡¶5[ŠN 3yÌkNÜ÷ä$Iæ‹é÷ncßs3ˆi`à[ì÷C¶,° ƒ<1Ó8âû“‰.>*μœðÓn<‡"Ù÷yâëŸÞS­Úè·ÆQÙ½5ª]Ô !#Á†ŠºÜ¤Zº-µÍØM»LËÕ“NÛn¿ˆqã„ÖŠ°EE¶I/ε×_†ì{ŵ YUHzà Ÿ=ÏŒãåy~á$|K Z>>ÛŠ=Þ,÷Iz{:t FĬ‰gÎ!0ŽöTgh(ÞÇ+í¶åžÚ·ÝõîåÒÂÀ(êTë’J¿eáýZ›`ƒœ}<Í»M=hQOôSâΞ)sõtž"ÁÔ7öÛQØnø G¨xâ|˜Pàû¤´Örœ†¹ Tܪnʆ—d håQc 6s¾±ç>,Mxg·|Ä*ÑS!øì noç ‹ûÆTAS…ªÉ·§D4%ÆÝÁLÆgûLJª ¦­ðÂGàj¾õnÜx°êK"æÃȽoÏ·ÕÑFÚäTà þå;’µM1>™½éOÁíßÆÑšŸ'RâÎÊ„…&y¹a±_™:·1îvÓ±»~î‰çzÌ|Þê˜7º>w3Ñ_P?cH+bmÉ'%ç; ~^_Â-øtfû©Én@£”SG\0ëà+‹ÅYNg•×y—¨]¬0²ðqI‚ â5˜Ò Hã2“&ÖÊÒæ§Ø€òR‹+Ùv»´MœÒdÒ(Ê­3(WÍù»¢Œ–‡`ØySM´+d8 r|Ãÿ+ ì8áS Ä1•YC2Ýè¹áHػϕÈYòƒy Ýèá)AèP·-ñaP…1`ÿ Ǿâ"œ}·”bC™uÐë#¢ª&ÙäP’ÀÔ Û›ÏÈ ¯™#Q´^ew#äö¢ž"+¢ýÛ>þôs´½i‚fVzàþótÛÚTÑTý*ÂÓUJÙÖ;ÁªuÝ6¸ôwÚœOì¢/~jÚ‚ÎtÒ/4{: =ÉC´—6ÃÍï½k>ýM@>+7ê`ÅÌã NÏRãhš/æMáŠÆeNi2=z–Õ3è+͸·Û&!PwÜ’Ù3Í4I3’;¥DŒ˜‹VQ©‹†&ÎëÊyØ;j´xßhÛYåëcË Yò¤{@µf[ IÐåÖV‹vÁý¼]“…Ø]øKŠÛA0ÃØlŒÕ{—F|ݸO;- ‡ï™C©Ê–ç›ØÔƒ²ßà ŸŒ‡»@H1*q _ä`7;ài†°Óú’'UU(ÙS­†øD؃ŽYXx|â­F9æC݆+K$´´ï¡¯!Û‚ÓŽEƒ1lN9gŠ890ßù¤9³¥ùÍûo#}ß5ɵ¯ŠJ„cï^ë "2MLa9wPÃ\Oö+Ð_²ê܂Λ£©‰.z'VÝ@«Hèi·râ™v Yr²Èm×½`x^â”A‹öJa÷ÈYš—çÜPéý¢« jXµgRµìXí¿Hm1hkK±¼ãj‘J¥j7ž¤°·7rݹ÷Bø"ó‚O4£´ªŸ%weÔ¡,üËÌ{–ê3xwE›ñ¹Í;ø0¹Õ÷Íï…•%$\2oöýN]ΈëÑ›‚·`¯”ž\G…º®ýéGw§I6±æM=»{ñIH•³K‚ÀÁžçÚî˜ Il8ô›‡Éª,¹tQ)µC!ŒáR¿6=™ê–†\¢ÒÅ‹Jƒa~äÅcÚdâ6aïœÇn%/YŽ Ú8r&­¬j¡4Rú]Á˜@Ÿ,VŽ¥rÁ­x«Ø"~ô ômÈ2Â_<ðvït®É%çÌ7?år·o¬ZdIü×`l2ØCëšìùà €Ä=ìÉò?÷tEý·5§ÓMµb3>«ª]î²É{6! ܽ›}ÜNµ†À½c†©æ"¸û '“÷=U²‘‰‘KÍ© 9êü”¢²ÜËà›6ÿ Í`záš3nݵo0åÐÚnËÌèöuÄÇØɾŒµÝ »¢`i¹$d¢“𠺛R2Â(ÑŒõ«c?ˆøMUš­"°N%"|çû¡ Èüßy\ÛhAPßräŸRõs!ƒÀNß" s¶ýC27÷-×`^¤²°‹<ص‘,K!ƒyCaö‚¿‘™”„tDûFOFÂ;®fãé¹+Ë/¥I·ôv8‘A8ŸøÆB ¾¯UÚ­×iÖ[È¥ŸBqÆv<ø5HF[ûZ»~•²Æã½2ãb%ãw‹ëNkÉÆ•=@ªV­Ûy›Ã‹®¤‡ âõil¨êï™ûH« ý—ˆ5eË5â×ÌÍÈÒÖo­P:u7ñÞn›Þž\·uŸæÜlâЃ&‚µ`k{ga$«OtW³_Š½²Ãö—‚ŽòEic†¿AËH-¾ÁÀ]ÓÀç„é¹ò—Á Ž¦O·rÛ¯*iìñ i±ø…sÇ^È5"dÏ@ícñ$L›ñýZ+ø.†Î°>€©Qüê0þaªÓ¤pšv '|Ÿ²>ÍœE¿&è-³ºÎú¸†Hëm“YÞ¯ËÐ-¤Š¸âçrÂ`–(9Q’Ö “!F—mÜù ¯ö­€Ý[…–®?–ÓWÒÕ¥¶ªžiä8Àîb€|4…íÖˆ)ˆ•?ØŠHÁãé¼+h49QéR¢ª¬¤àu3ÖTÔC©ÓòêaÝÄ~ìägT¹¤%\•––P¤ç¬ö¬ã¨Ar„Ö:«!’–{T}ÿ¼'¤‚=ZlZ ÛsÓµ4g=A¹ü̓êv0>3Ÿ¥ºU¥¹þ©) ¬‚™i‡ý39›_UºòWõµ‡B…‘ÑåFàÜÙR-—ŸŽîwl'K?­ŸLN+ƒJEáZ!{Ïjª@ÒQoçßF-Í(hVššT­p aìg±6Ü‚ã¹j¶«@ßbÌIiŒÅó/à»L;õµS\ `ÙŒí¨㪩tSk„^Þ-(õå¼ÞÔ=‘*LÚ"j0ŽØ.xQ!úÐøMrZœG]sÌE'„ã.ÜbBû’k;’Ïa–áòBvÀ{ð=w½|°úªˆÎü™Œ#²Ð °ü3koÕ3\ ù󯉰³„Oš=­¥6ŽÅÂçêaýòѾ ÊXËOÜÓSÂÔ¯f¹ j Æê¿6$äÅ'‡ÅI¬§+¸„Î0†Ç=ïÿüÁñÛ‚p¯(U·.µFŽ¢¹VÒY‘ö Ý †g<]£lÁlG—9d«.©ÆRXlDpÔ£Í#žà!o¸á}MºSp„b>à ÝL‹¼>â’Úâ6~üV&šè£™¶Y¹œ»³mÓ G> §2~º”îØÝñqôw®N~ñæT•qühO6-¡:Öc¡oÔ¯Ð}1Pé‰8Éš2„ÞåL˜ŸJ›’Ñܽé…T›ì‚“¼;q"Á])¢ÌrÌ ÌÖjy4Ü0.Eõ£CæóÓÎé#Ëc˜®1Ç¿ÒG—P¸OÇÛÛñTáV˜‹d²OŒ\c.Z­vë%fʦv=qí$™ðÉ¥Ÿo‚@Øvfp€Áté»@¿8ÕM°»…öæîy;>4ó*ô\¢•ð,}“ÛÒH‹š|ž¯ü¿r¾HI"ª·ßž…Ê* à ª_«Ç ƒoéÆúØ ú\mûÝZnÚ§ÔW¯~Vïj6@(D¡¹\òûøÜw)ú› À»Ò_8*Ÿ˜©ÄJîð¥é›èW’Ýñ3è‘žÊ%zš¨Q¬ëõí0Ä¥‹Oè¤m~ù¶N—î“£¹¦¾âlЇW"¡ÏVî_8îkÖ¡p4ÆÂJ´Ð Í°9]à<=Õ4[Ê T^¢ÏªEOfîvc颈Ï-§Çeˆy´Z÷q÷Âß ÚQŽ8æ‚gøWݜݠð@A¨¬$îOˆ*æÖÐ8îWHH‹sMœ³ýÜ^«2ÞK²"ë9LÌÃb æ -ÊÖøп ¤±Ñùq 3êÊøpÞÏ6t 2ï¬Pœ­Ë1|Ⳟ$íwTµ”Ôn¹kYälìÄÒiAîyû‰*€9]«ÈíB¾-¾T\€ãfÐò6k] ÷·ì+žüÈ›láfq&üÊ‚ÊY²2e²`„ÖnŸ÷€|‰,–k1t»ìzÓˆCÓâí«)׆Ãuþ2RpE+¿’¸7"`]GBO?‹œNtû;büºg–9®Ø‰/ ™¾ƒEZg¬÷(Ç ·7-!Eo ëÄ2fÆ„b6FaÇ3Û«¹Òîëvc1RF÷\ æÕ·E„ÎBu§Jx4¥w{vÅþpYk1ò¹¥d-Üõa‘Dö>j÷J”·éh µ¢%a¡ðˆº«õQ g½Ò“¨vP¯9µj­ö÷+´äBŠ#ÔKrØRsC—¶á‹y¶¥%^!¡Põ1œdŸ‰VûOØ Ä¯£ï¯BXÆé} ’DÇôcTW+Ë~ŒeŽ9ã•„²û±T¥"µäñZ`sËÚðÄ.?ÙQ»I#WÃ*a$ÞN£¸ Ú øL#A$gœË\eR1VœJ³‘§ÍéU Ö¢aIÞg°¤ôÅ0 _t72öÙ|Uð '0,æÑåÓè¶ôEDÒ%RM€ðN[ûÖž†ý=ÍA‡;#D'߆ŠµˆfÛSƒÉë1D_Øt©ç;ÆÅg¼{¨´t Ì,û\ûnµ#}C:2̾Ëp²öþˆ¨¨Ôç·õ‡fWµEËU6=NÒuqþ,Ú÷Å…õ&4Ø=%kÐOö˜Ë×ï9vaU: Ðm%ü´_ésÀ¤m6'¯¯èÜÊÛ³ÝVvÉõ!¶»lÓ™ÖwÉÌKmä^sVô`I©q„7ÍÃan½~I!'g˜çš¸ÞÚº˜ õ£n•wÎŒQ¿wœ—zÔžd4ý~‰uÕq?bjàILMGmTHøªl ;ŠóÁ<2Т05ô#Aa–»UOKá˜_{; n_±ÇX èú0\t® ²tÅûë­PQ‹5ìyҚ欦Ìââ—p¶1 qòŽöãö˜¾,‹Ø/‰¾  ëP ©hÃó‚×ÒpÔ™c9ÎÏý¨mùM;—Ùº™69lMô£íæ÷w¶,dÿA@8W\ôÀôêГe…ãIÚ¯pyŽGÎÖªÔ£ˆuÖ5[­Z}"Y¤9u‰PÁ]¶LÚ£B2&!x²²^aHØW¢Ù ÿo ~JƒÃLÍPÚT ê¡i§gð• $—ÚÓxlËt!£ÍQA0³õg0öœj¿êÜÚÕ {'ð˜»E øÈ7휞¿¶ÖˆÓ+€€¸Õß… S¬±ÆðùÆåjÝL ;qn£´[ýþèdjû岨9CðVeoYϯ‰Ê¡¯kaÈêÛÓÈÛäŸÃÒ†ò¸ÏŽe‡à–f-ƒ¼ÂÐa¡‡é5VuÇâ ¦7À~'¬ELF[H5p7&·ý=¨7'_ª±Ðçí”Ê.ÆÀj%™°ÅÊÌÎo˜Á6zä¥Üã0¢r“–%>6F¢ÇR|ŽUrÚ¡ÛR(¸%´“±V–‚:ùªŠ«¡¯fÛ×ú^_ª´]Lâé ~mólyÒ¿ÒZΛ‡Íò9Çç›jŽ«Èt#LëCc¬÷OéqÎ&º—ëý—7:  Ö1L1&DÖ®Åõ/µ¾E`sÕël:Õ°z­hÛû$ÁŸÅ{ÓªdóÛ¶¹£éh Ÿ}% ÂU`Œ=ºÖñE½pÌ°¿x+ëG͸‚&ÖØ¡·øa¡¼l'©t%òná•ÉîŠÉ aÉRÙuV÷üÒ&ÙîëŽ_•‘þ ¥2¸³. î÷ìºwnÎ€Ï œeåàc³4X›lK:è™ÚoFTÝGQìQEæô‹ƒµ¬Ty 7Ÿï£k ".ë"VÇ‚¯bÆçSH/ŽW·!£Î&x§GåéwTÜ{ÝBǘH×Ç©æâFØýº –̾íæÏ"Ìû¨Ó‡ôÃÀž•³ÔCÇ|CKàúºàeõ®™mÅÎ{ôâ .z}àvRjâ=è/COybD+ñ“į=Í-R'Ð[jþtôXà©Ò³‘ É~¨²©Zâ ËÝÆ|®ÇóaÄí÷¿Õ ‘NÜK‡´Óž+ož# êö ÊÕ)&ËÙ»ÙÃ)8M:‚°ˆÛq¯‚{¤ŠÈ"ëwñI%¶Ô–«Ã<•F™õ~‰™öwËSOSBH·*z±Æ§ßäÜz©¡†Æ2xè¹ÃªߤTÈ ~Ç)·‡S󦌠új }ê«ïñV)‘ÕT` QæÔ Ú"Üýlý9„Ö ´&Zι»d®ã3Kj—.\hJëùy{–z¯Sã£Æw‚ÔX»4™kšéjÌAAóFg~˜ö¾ÆÅFJÔ‹_‡¹oa)Š)­›Ü!©fo-u´ü)q‘ng㸆;,ÿ@ÿ!bqîÉwN¥U¡}4¶>ÞI<ü­hR¯ý>ÂIOž3Ò–òÝ(ŸˆÒO3òSî LAüQŸ!/•%5XòŠô«§>Kå ž€©ËÙ¬ùBµnw¯{žW_¥àvùË-qš}÷@ž±]1ë§;Zn{Û˜žÃ?í\û%—(‡JIcÂòðHe|Å ´!e¤“6¤Þݘ¾³dQÉØ’“P|ZÄ\ªXǾ‡å—ùMuK}ø9¾º* BŽŸùõ5Î]¥ŠWÌü³š$›9ã}k©s’3ÆÁšx'dÚº,Ép­½5Rt5zq€oìÔ‰•‹k-Þ]¯q´38ÄÑ»Ž4“µM½™ÖÅk`~Š]16UU±YÏOi$p§o÷H¼ŽtLº¦J(ˆÃH{Û×6švHY›ºËL½­>]ïj[›pYù”’ñá¦>âƒÃ5"¬ên[y)C‚c¤A€`óUÐØæW1Í)©¶˜ôs©É¬ûs¼,X×ÖÞ[$û*êtXw¼ÖŽ†ùXŽlwƒ¦0äI‡;ÜWuÏyè­&Ý®˜¦ºþœŽbŠªü¤9þÉÙòûn;yªb”È8³©$T-¸}´úGä—}}|½W!<ÑÙ»NÄéÍ×g¬q}¶ÛÛ­* 郞T‹ ­V`ø¡M†×'¥ÂÜu‚âÝ0—M5¸ûÙèjø@“Y?ªH„Ò)/¦FOè#' µäœýП.¦ä~Öoú²%>®•8òÀgU=Îl6cy„vÈ‘,Sä\•9â0wé‰ceäíR9©÷j,âtÜâÛ{‚f/X6–…i-j$ }ó­ùÑ¢»Kðâ#/†ËO1³ZôâÞ¡ÓµdG™,KÊ.â\Ñ?ÌfÜãê…òCSB\20úZĵ›K{…ò"³èrtç7Êܽë<9nx"Üf^±Bx]†·¶l•¾xRŠð‹w½Oº÷|³oºåëÞ²x¯\2!½° IÎÉÃ#Ô6e³‡\ER¨Æ 77KÏ`é¥r’DèÚ–Îzÿœa¥’ÂÅ‹ÕlÈ+ìgûÅGªY§›?_S£çÌBxf–m `g÷-9å\˜§d€ 7ô »0%U:"âå‰"kÏ*Leý,¿­b§|Ί1RdW釔Ù‘µIG–¬ëJæbÂ,.DTI?°C¿yÖkË8ôÝæZž©x]Ý•|3ÇR&gÿ ð,.ÌHßPäfô«¿éøL0¿P¼ÀÙ‹º¡a_ +wxpZ\0weªðEœ÷}UPq½µb¥5É iƒŽm’tmMâFLlïæÍ|áòQïBúÕ(¯QÝ%çÍ^a²%W¾®ÑÈlõ×n`C„^{<~—2Õ-‘î§Ê+Sl‘qûæ §c_ÎmæÉiD öTsn°øLœ|‡æ5Ññ…(à׺À§£.©Àíß‘àîˆ!.1ܧzØTãÝ ¿sû¦8ïÆzÝb?É´@Òuºƒ^Er<¿[̽4kµhë4hrzñ;£ccÁ¹õAEŠË|Mà‰›úÖn­•îµóì‡åD?â6@pC×Úi•ëÁ*ñÖ-Ô¾Ó®w=-Â)‘næ®äú¢Ò ¤’n. šßm»&=¿ ð—0VE“À£ ¦xqº—Íý9$óÍ &ü0‡NOÏÈ=²|#moŽ{䉳ÍH@ÆoB¾ œHEç)¸ælÙkêaE†Å–«ošZÿ^¥[±1ð`«+ôš¡sì>K+š6¡p|ÔÕ®ÁñDmžX„ïõg&ÉyÄ°žl!›¯˜f­QÁe<Ì2(°È«w,‡PMxÛûˆþ»q8S®Òeȵ'®ßÖÈÏè$ª #— ÞøU¢odJ*Èf4#>‡2Ï¥ˆá®þéfÌOV…Ù¹aè ßX#¯ÕáP¨ß8>PÕüZ ŸŒ/Û¬Àé\Ýz6È1Ć–z/Œûr;9nâ¢Nμ)u#Ø‚6ÿit ey­1S¹[íhHý¤PœzzQ›Þ2(XZôŠ³n÷so(}Òöyc‡du˜½k#±h›†òW#bm(j9-.´v?¬]Ù³¾¬;K¾ ¤–Ñ—>ÊÇ ÌXú•KmÎ-˜­Å›¦KU:,;qn†‚Ì;L¦êÞt ¸Í`[ToVÀ9iVÉVàÂM*Të*÷‹CQhfàŒÍËØdòÍž£µÜƒâºÍ`<CS"¦è|:¨nAÄð^ÕŠ¤ÿ1ì– d^µšG0‹?•‹¦ð Ó w‰á‰¥x'±’`±Z¤¯é ošÔY0k‡3"g©†bÄvY"´ÎÌh¯KË“ûœ‰,Ä54¨d€#š4ó“c'Üa (^M¥ ¼YLKZÙ…··aS+„  IAÒG´ÙÝs̃“¿˜ê$ pÔ¯ 3"gs ö«âh@·Ç=è\=¬Âèq-Oé?_‹àS2ÜЪv&–YdÓL„­ü>ÚU@g¿××C<-¬ÐŒÀd~99 æ½wêv°‚WA‡KÙŸÅ``œï¨µñ™%cÒ$ºÔ'©Ø>Pe|7[%ŸòÆæÔMÂï.e²ËåPUyð¦ìÛ"FLÝj×$p:bdqû¤ E‘¡©¾i}…õ¯•Û+a)RÅ׿`7W“ùV…Ñfuô+V]£lÇ'ð®·56ÿæ1!B‘ >r ¥!ÇÓƾY,ñØHpåàxîûYän<)ýæ½°s2+ß9!(M™ìXJõ<ð1zÕÁÜ «n@ßÅ'Åû,ñØùßTýÐ94Š5M_Ú÷!²×K.CaXûêyŒ§±.š ß5~õŠÕÔ{1uzpíSAyûÀOõj]éí–*h¬îÊ]kD¸R ŒhÕ èX;Â)´5;–“{åy6ë[Dwô·¤E;Ô•í¸îŒy]\ÊwÈ-(à‘sÐxPnéõŽ8À(½îã{ÛDT}ž{M§U©­ GGt4o,¶˜ó™µïÄÉp·ÉÂçRÛTÜ Œ”‚&Í÷—² ȲS(a€‘ê(à=ÿj€bYbá(1jü&Á£8ÿÙ¥ç­#“nÉÒHåÅhÔmSFÿdyUVf[ÿÀœï epÀ#“¡›®8 C`ƒ˜Jݪdª2—>ƒì‰éƒÂ1ñÝ49ÖCºÔýú=ŠÇðè×½R”ˆ˜7,1! ,ça¬ì<›ÑnAæm&ø†ll΀§ÒÙœ ̱¬åÊ@[g¥@E¥W¹{á®ìï)…vüthͲðú6d±0[Ù»Qº·â™ôŽzÇùXÖd‹®ÖErûƒ½§HžX9ÒwKʽÜæ@`œßµŒ>ñ À‚:.ª_}f;Àcq\pyx1*ÙÏ M“ ïá‹1z$I% }íYï(ÿ’Ì„­ÏO\Œò\1µ™*¹cùòÜÑ$Rü¶ô¸ñ6q |»ùu³Ï5äpUdr/{˜[(ÿH3­ŽP[.û·ØziQÄX¡î/á¹QáŽhÏLa¹{ÚWÓ\ï[œx <Çh(òë燠Ë+ùÞ¯7 _öÏ*¥O•½!x÷]³g !Ï8~|–¢ÍnFœ"¦36éâÐÕ¥î½<.íšÇŸK–f¡|ö¬\GØÖÚ~´fiù˜Ëí»þa“±[ VäË—.i8ÜWòëKvM¶oÖEL®¡*kªü1ÞœJ0S±Ànlz6PÏHƒ}ϯ$ŠENy=°™7}Áº©IË>ù|>)+GgðhClUØ‚8"‡› ¼ª V„ˆüÆ1/ ±jù»ÿeQyXY9’):Þ‹ÛÍI* ¨±Hì£ôcLäæé„ØRVÁš'žÚh¤{ºçò{Ày. »Ž"î‚. ·ô§ž¥AÍœ¶Ç¹vŠ|¥^„µÅüݱPÓü,•c­b±©~t#Ô‡âí· ÏÝüÔ8½ìÆÕé ÷h,½T+ P›6l°ºÕwÖ;O=" >j=Í߯Կö Ò -8‘7“IÕ~ðêx$šëpõé´$£O°;¿…&¥èN ï\‰±h¥L õ›Œð§’ˆ«X2ÍS~b±#=èfw3fÆs $gâö1y¨0`Ž©YYľX8Òyúý¤_;Øü¬¬0¯`ºávD·M ‘îÌW`øŽ¼?¢þó§îk9Tût¯ÀÈY2-*Ë>»Aoðù@.RâGUgIV~‰H_¤ÙPhzð`ëÐRôrÒ]¸ˆÀ$¬ŽsCRÎ G0ËÃÂp¦¢´!ËÑ6XÝ‘yufÍ÷ÔW‚‰µêd&øHÌNqJ±ÅîGgÙ«x!8O7£¸­ß‘£|C·ˆŽê‹¥Æìß´YEÏlJÜDª¦Ö'{z•­cßYa˜¯»äC£ªìÚ¸ÕS¿èAàx¸—~ûGŽ" ³× ›¢’ä{ÉlTÆòˆÅ.å»%±­ªÞÙfíò»g´_Rø©¨¶ ':@’¹|yÙìq7’–¶¸þ¸½â¸NÀ(3 ‹ ¿[õDµK9%ÊBÉí‚ŽÙÁjYØ“zŽ†AÕéà2ŸYÏ>u稃æXõNdod†¯éü¼¶-DïtÓlC²®$Q˜ŸÈ{ ·ƒ´à›l=¹¼/-Ûùàl$ 6¹„— »› ۠جê¯X Ûôù€ìšƒ{$5šM«(§ð “ÊŒ†¥êò<Îjw5}²¼ŽñªÄÚòÌ%ãÉã)ªˆ}Œ‹‡Ž¹ÝxŽ{Á\ÚË«#ò $¹c¸<©À¬xl®< åÒºœÏªJ¥°‹9Y€ìD¨h&TÖ¦#g?̶g0lêÞü o”ˉJÒPÇ–Nì'0eOÓ2»ÞUQ±›?Ÿùß ºiïR0¶Q‡¡;)“ktxÏ™šQ¼Âÿòü=”ÃĤ.¥/Š`d#9u嶃õÕý(¶þŒj$Ê æÞ¥ÿhìz´ã"ÕÄrj‘Æ{vÄÛä— Ž¼ÇIͼÚ:&ÕÐöB4C¼»g†\ Ñ»”±ÝÚôO„§ý/*û‰–’MØدXR|h£C˜Ê-u¤à‹¬4¸µÜ¢vt+GEZÿ+Ï)0ÖºŠ ÷[±“Ì^ßÐc½%ã4$ÕiA­]/7¢Y ÑЦԾÓGÓ•:ñƒ|qK®ÿϾzÍa«ZÊ/e©òa’ uÇaýÅrXM[ïë†ÀkŒ—I˜ã>";–- —+q‘ízz+¹I(ð÷ûû„ï0¬dh9¿sbŠ—i¬ûÉ÷š=§^°’¨Ís÷Ìm‚WÙ¾¡f‘ØÏBnø(yõ0ª¾'˜g_qRÅ+Ðú£î¬ú+  Ùd^¯&©á¥™`ÕoŽ•—jâøµÌ?)„FÐÁ(R;sоäÄ•oÁÕÏZ­µƒ 3u†ižßÞ´\ümqm¦¢¿ÞxGmö¹]Éýœôˆœ·+DÖ‹{´šfãO‹‹,Ï~;1~·ã2[-ß>±˜Ã@7Cya ﻽¯§ËÔˆXúˆQ—`ƒÇÆWØÎ"ä•y™[(†UçÏ+ê3‘ÅÄ"ì²ãöÌ¡ GlÙGm;¼¹í³'+`ÿßKŠyÅ„u¨¿Á¿Dé1ŽËØ)Š±žþ€ú¦©KLwͬ†Yú\CéW„ŽB› jsŠgý·¿Ø/XÎRÙ?U«4ÒmýÅ/O*6Þ/‹ý*þì$Ž› $Ë´á‘É} c<ñÃÔ¹ãËÊ«Úìú›š!7»°Œ…Íà[?QA}yîЯš2¶ø´¢‚3qD<`½þQΙðªd%8æ“wý:ŠÎK’`Jœõ¢YÎò¤ûgÙYÕWQÇlžB‘ [vØ,ŠiØC d^¦¼Ëkî3ÓìFh gÐ0u ^ñ£ ËS×~Brö¡MO¹ë,eŠD„Ô:¸Ý´úþÉ]®g[Öež*áÙF2i‘Ö5{@8ãL¼ÚÓaþŒ€Qc¦•}æªj¶åÜí6ñ3]dšsjâ~æ# QØtí×ÐW·ŸNým¥å¾œpmf¢Ñ³r™Ò?¯ñì'(½ß_¯š60^6¢ý |Ž_¶¨ âQ!«6¤ ‡±,ýñYÕ/ØE_æû;‰ëÌG—Y@øÎ5möWCE>ÏvçÔ®Ò¶:îó«7g3ú¿”}ZÓ*B'¨(œì,½ú›ã øÃâíû¸R/êÒ¾ó«DQìPµ_Z·Ì;¹Ž¶]hëtÕ|¥b/ì+Š£1U6&ï¦î)šOÜ°!w²oØ>!Dš÷kˆ°e½´jÊXbyëóÇ9Í_¦…ÅI’º²r½¥Éôþ÷9©<²ù;‚BO*]A©çÔýKÅob Œt³÷ns1Á8Æ\²¼£F —«ê•–@v.»!ºìE1_I1î- èXvk‡ãh„Ulv¿ÞÙí‡ÙOE\¢K6Ø’cBpÉAîÓ ¦,öدŒÁƒ=qtw*TΨ¥?µÙ•‚_DšJÏ´®cCŒÃSè>{xëe`¸0db ¾±Õnïðš¾4Ñ™àrîGæ‚ÀÓ£R®7j-:eµ®¡ÇN+ï+w s‚mÅÖÀåýÏ°ÓºÃ{:ûUÝD幋‘‘­d2¿·ì&Œ¸X0^$þêô¿v.= WhcÚÉ?‚îƒ 9_5OF¸ËPè€sÒܺïC…'iÊ ¿NˆøuL:¾ZÒsaæyµ+lš–Á„çêjó–åϙܔ{,’·›Š :o"Rp¿ünsá--6æöoÚèIf×)w=5ï¸ïhôQªØÈ©µ$öîGŸk¡þÌ`wùÓMÊ¢áWZh Õ‰‘Í{`隧>©¿—Dò.ÛÀ,~¯ûÄXj•¸ÉS-‹ÀMóW‡°]Ÿü'ü7÷E£˜@‰m;W‹"”œ‚a7Öñ@ÁÏn)éÞæì«Ì¡I{“¼ÓJ´5­ãÄiÙ¥D;Ø轌ßa޻ͬ¤a”ÄÆQŠ;ƒéʵ/‘ô5ÍDu ,'YgH™ýßå¨É’ÌeÚr_D*¨ù#Ç,añ뵃à~–öͦ{y¸Ú‘ýXzÏÃ,³}#ΩóÁ`ƒrt<ýq°†öæÎX9懰6gT¢ѨcýÚuèæ7Š­JÒé-ŒŒ‚9 S†”õèëÓ‰¨‘ÆØÍpÅåÑc‡ì‚â žph™½+Ó7ÇÊó"[+wÙ"ÉóãÌÎÚÐPµç#`&øÅMo5ËùM sP€aÄü·Ð_ÿxáAÑîÕ” Âó³à†(̸åA³iN"Š%IˆîÆ ÛÐbLh£5(ª#œ~5¹¦AÖ™Q0´t‡èí%æ  ãY¤×öäÔ‡Ô«ßµ—oJærDõ·×œ*‰†1`yW¿§]Å=pƒSžyYcrŠ Ï]*ôýÁXîH[šd™ ~žÁVjë 5O×S‹`öžÀ ØÓ¶ù€Ž„†P馦³|ÑéJömhX)½1äs a⣤”C_Á“Ñ‹»4ÙUô¢$B®¸Fæ‘]Uå˜Íš:i¾7ÁÑÞg¨¢Œþ°÷°ÕcÜL‹¿ÈûL¡<>GÌçN•µËnÿ|HlÚs~ƒÔüÑ1ëzé@m{öß i×Ë`BLMÝJN,ÁEÔ•È.Š$ë´p±‹ã° ¡øîƬRr–‰ÑÚNqˆLø€fWóûxŽžŒ¡'n`¥”xìjV+&ý«vA‹Üq°ZN˜ñu?·i•›§¡ô³Õ¯á¢~þI F%ã¦-Zö –ʼnŸ¨skÚlÕSør¡Éݲ‹Éc‘HHò¥tþK¥l¨¡Vr…Çã IDy2z:Ô¢àÏ(ÅPLr mkì+ÒŸW2c5©·YhÞ£f ÝmzÊã C¹M…IÒ»§Ž•Šd⇣†¡ýNþmhmWýÞªÛœå_4­Ÿ:ùèx·ïíÏûdàÓ}擶œëY§ä¢Á iÚ™ž¢íM-ë½Xê,±œûi¹õº¦á÷Å1eOÎ&‘:˼ó+ó;5 ™^…ìãj˜e±¹jtkËÕñºÓU\U-.@0Ž¼UïÐÛȵo¿T„œ=®…“ˆº7‹xnÞ“ìRsÝH»Á‡àò¤¾E¼o!177ÿycø{2M¼v°¨ë@KGš4¿¼(ŒÝ$’ñµ§O ÜÖ 4 ïÔ²¸h rÝ\œå´¡m68†oX¿é.¥ç’‡:Å\hZ÷ÌCgÆ{ú\þ´9q¡Pkäï…Æ e£k-ÐiB³ "eŸ¤W—Öž†'Ýqy¦8¶C 7®NQ2_ĵ‰ñÿNJ\N­0æzdqÇ“ð¾¨ÒRⵯ2±¸Óí{ÿÖw§€Åyoɬ ˜Àhd Ðb.*ûÚ?FM§–É—êãÅÐР\\‰†I}ìß^›±‰Ã=—+ 'ë:wÈ·µ÷æ‘&a”7àˆ‚FÇSœ`mãˆCƒÄ$µ<ƒÍç~šòU Ñ·þ?náM+/†\9,Ú€jµ%kš:À©ŠQlÖÕà/S¤“°æúR÷ÿ…‘#GÔ%2iÅ~¼ÝÙ¨´H8»Í±ùÖ'jâDØB'u¡ìÀùËxˆêÛåö¨¼ÛkáÓ^É=¤:†…¬±ùr{Å|™*]Ñ~r¦[OX Úh€_ø÷©Ñ}óÅ\”Kgu‹¬aaÂ_t«A§q˜>“Ÿ áe$Þ½¥¬‰]NE;ÄÎZªÔJ…©Æj2³*0\óVÿâq¾•ÞV{Ñw Ýá6 YáSÁž¹ù³Fž×¼N_`{‰¡§xø"œb<¸¶båÅ™þP‡\äí¡‡{+m+/ö §lûRËj=º#wU6gÍJ‘A¸\Sýþ(Î~K§ÈÂ0ßwŒ›ú‘7î¥ÞóÁÅÞ|¨´²WAÖ§±ü%­ù‡þ‹OŽ¸*^²@t;¾&£ Û¾R/[dÔö=¶5ÁDA3÷R0«ôƒ±ûØEj 5úZ‡Vr"»k¸@`–-’Ý“‹ôêXD™¨Ë¾ÏnýJãa)¢d~òKq®b»ÊjíÙÓÚúj¢X†AgŠ„l JÏÓP|…Ö÷}]ôKJñ»Ô%5—¦µå·&?cÑå¾7éÝ‚­.w[øÒr ÛßGQ@̓}¡Ú²÷c%›¦_vF–‚b™Ôµ½3oa6ã(ïWø´/¬ÚFP &èü¼ª-×^Ö½s¯pjÀá£9QfP¢DQFCª³5Ù\ìåŽM¹” ye†»ÆÎx‚žf\TF!õy§†Qãú¤-‘Ö"¢Ž#VÓ'%j6 <ÍJ âÄ_‚Mìph3$ªüšÿ’g[ÈÞ Ñž_ dÁ·*Žk·ÏEúð£+öaîÔÜîê1³îƒÀµ‚Rªýš‹[Œ®‹X¾L¾ùàUsÒ g€ƒÿÊ1âg¶¥›Â>ÖKzíT)¦îtOqì «#Ÿ‡œ6  µ§íؘí&Ôæ¶èQ—ͺÄ:u­T&lgEòýÇîÚÜhSó'§\lŠð€h¸ì½Ö§u p)¿^hTëUFàïÃõ,Eà…Óm˦©ÅpÖgŽ' ]®Âè‰À–ò<Ý:•*|š u3`ÀÄë@}3³+¾ „“­%&–\U Ǫô„ºÚƤaW¹ÍM±CøÌsVÞ Ñqóu°(}̺¬R{ ¥ÊÌqË Tå3‚?ƱÀ¯ë`ø JhÉ×Æ ±ÕÅÑ_üΙO2¡*¿TÇ£Üï/;æó.r63Ï—Îèç|8[äueÝ)ÏïÇ–NniVëte éjŽy»s Pòâ±ØÐä+Àí?ïŽÙõa î“vÀ’ý ÏÅÕ*?l¾Zä€X5Ø<Ôôny®|Pzž’{@uÎó$'§Ï0VŸIïg'@û;ÒöpÆÏn 'ˆÅ­Wxù'‘Xçˆa|À>íK æaÕI½2öâÅ`OÑ^Çì8ŒI·¤#bÈ6}¯ -<¯!&ÇáM Jw—M!' ³›÷H–g‹×[×g6±ßV¾–½Âyo_ô*%;íæÐrhæèÙ¾Bà:(@ná.Á%þpŒÎäÏRŒ,-ýG°^\E;ÉÑ;?1ÚV\kÃÉŽ±Öع5rú™¡»sБL«E¸[.®Fk§?6‘:Æ­Þ¹¾Âî›öFbirRê_µë¥j0ä”Û˜ÁbvŠd.x–fp±[VÆ{/ØMAlËÚt¯£,[hÃë nàÙ$oþºÛ¼+X†žÀØ!uý€qˆèöe“‚³fùËœÿ‰ùš1@¡->×Ðß/Pš÷(ײ_Z 5´(pVÒô…ç+üe´(mЂˆêÕLwÍž{Ö[ŸVú{Šiõ¬ªå§üZÅz#¤—E>¯]v)Þ rò$Ü;㹕› …Uú-+“;®š-®E:ËgZûþ•fº sü2ã6Ô#íô•ªKz0Æc²'ÛÏœNõéál:žxq?ŽÅJ€8=Æ@ê=V´ . h$ÕóÀ¡dq—¡‚å À’|·®Ÿ¾[:ÜŒì¾#HÀ$iEa¸<©Û¨%lœ£ÚÐòºˆÐi©áãÌ(Ÿ¼œ‰ f‹¼êØʨ“bû_„Žu;:« gRèþpþˆs-PŒ\qß…iè EÏzó`B ÄùM Ü)åW‡ 5è·§T«ÉÉ<¼ZÈ‘7÷aóˆ|'}¤¤_ävÑì ua±–Zµ¿/EçÊ–óH@UrðJ^´(–Xý>+üï¹Ïãé ±»ÉgàV:CÚ#Q¥´Ý€. Õ*EÇç_À¼¨†Jâ&©á!øhséµ»ï³h „ó¶3ëÐÞØ1Ò?¼âbIÙYö ­›¦²[áªGF_íãë›Áû¡¾!Çå¹ü±Ìæ݇¤S«‡äM.ïÐŒnèÖRÙ ,.Ú„‘‰"·¤Ä‰ïPìè2Ńk¦;"[ìNÒ‡9ZIVw3_K8&KLÈÝð£TËMX,ŸWoÛu¼J q>’ídÉù{§ø¾*æ“kÂÊvḏx¢ZÁFo'q:+â+ɹ.®ÿ9p)Šúî`%š`¶æ̧©Ø5‡äÔÕøÑš€´G>í¿`,5yÏW®s´šÝ'¨9¾ÆÁhÁ¾__y$›7Bev ü#˜\ÿT G\‘ó^ÝYdý‰tRyÀZ{Æ-wW¤;~Ñœ¶Él³s)ÆðWÈUZÝdA5êDkÐë” ÔêŠEwR º¶ôaÐåX†Ì¤n˶o†Ÿ6‘OãÕƒ«!û´$Gg¶Hu€2T ûñß`Œ¢n§K馢i¸.ø‘”=o×ÕyêÖê€ö¾ïƒÖÝ ÛÀD¼ °ˆå[ô‘æ“i¾Vþ|êZf,ÜgSžÂô—ãs¹”jŽ/©=¥È¸8Ä¡ONQwzÖ•bY³ÌÍ`F! ªÈß”ìë Yµ÷ßG!iïÃüÌç\×´`€–b9$ÛÃÙ­Ð7€1.2c§ p»<í^rµNAó‚äFq$WH õæœ&wÆöÊ]¦8âØ+å«q±¶ 2®q«ÖvyéšÎß‚ZF˜%ó䧼¬PÓ±œM©)ûì!·ÁL³¿PÚÚ§?×~}ˆßº±m§A²"ýLäSKl…í0÷Wh@”±ìƒÙFtÂ4É šv†ÏßfI”¢f€•ØuÞüñ„ ýG@r/ çŒ>™l\¸Ÿ _üs±DÓLÍ35 "¤ð7òBK{U—oVå¡Ñ¦cßJËÍ~»ˆð'>·~_Š]¬borª¥ôÎä”Òs >=4Âéý¹ˆd» Z–C@ž€ A*m:é*£q0¬1¨ö&×ÎD(™u«¿VVzU´4©¢N¥æSæ2•¤À”Ó‚üG§6 Th¾¡%꿯f^‹¸„ê¾Ò¸Ã–+«ÆùšbÏ`HVY€àF¸ÁÔ0ÊŒ³+£S­D2¿¸©„}“dn¤,mmf|÷Ô˜ú†òÜKIu‡&_Bfwõ¹|í±F!i_HŠ´‘0|±”ûµÉÜ]©cGˆPàçu·YÅ¢+’øñÔ…ÒÜ™ÕñÝ”$fl⃠Fä¼™g<õu!ýDä-ú(Å[èw[zKl^£/Ñ®h=¥Ó>€I•®Øª”¿^ÝPlÊ%àBJ¬ ¢1‰´ P­úhÀ^‹Ì2Õ¿, sÐóÎÐó–LŸ?÷ßC*õšwQŽÓé\-xŸGF˜+Äëz|+(*¥jG’Ãÿx&ŽžªUíùÑwé“G½ëO2Òïër©Kìšù.ߎ©IÄÀÎoÓ>y¿á æªæVÊ¥åñ¶}^³ª"ñÓ«$MOý.HFùFGAMœä:[%¸ã~Fº›"±s(Òc¾Þ›zú^Q}0þ‰®¡17TöB•ÖV²ÔÚ¹ëRñí„U™ÉÑîø¿;åé°t×òlý‰ÏPüu7Ø0ä=à><@³Ú`Ÿ)à€ÌŸÀÈ O¤GrØ÷#¹ì%ð¾þŨ3" Ê—ú?01ä»k}²ƒN¿(زž^¿î]Þ6x@KÍžSbƒÂ‹^Œ¥cm‡]ß_e`uÑÆ‘j¼pƒfÌ2MfòŽ<>mÎù=­1G.SUuzkørx[wâàÚd0ZªÂkìöBAГñ ÿ•ºÝº•Šø™444\P¿YRL½ùCAc¼­{uù>Á;ùårùé é„k*—¿“—…È«Ÿ‡ó»o–•L_ŽEœ}Ý R¬ ·²l{©ÔBã)'z!<2Té`mp”žCŸZ4Çoæ-(“œª¢;-){$UÀèº!\[„¾YŠêðˆNÕ$Õ{´_:¸5v@}Îlý'Õ‡¼HŠ£`.™H$c‰¯sM‰‘8Eä°ñÁÁ®×tF vò¿$—1QÈߨ¦ÂI DÊùƒ\ªj"}[v€÷ðNQ(-œ¤~=ü·°K„®ñ<´2 Qaû˜zŽçPHpfR3ÊÈ×òUr‰b£ÕG®sü»]ûe¶ªc¨®W–üáW¾y²g•c% ´¾+r»ª½gg`×?û_ž=Ï3.¡$»†ù®x?–kPo t³îó--—VöÄúÈ ÐÍÑ“¶ßÛá(åR1<âï{\è& ‚àFŽÞâÀøV‡§Ýv±7 órõݼÑV|ÌÝ*õ7šXZ°©£?4ne¢¿lNÇ‚¥[óª*öºÙL£±¤†½Gª÷k?á²°]~KÆpf¯7%–ö¢AÑf;ÂŒa ÛÊs7‘HÎÅ$ãJ¯Æ4&E-ÃXÉ|šªs’˜|nl¬ Á¥£mùˆý­?FÞ DÑ(Lï½üÔu ù¬ÄìÑOaK¶ÝÀ5½1LBá%G©’;Û/<ü Ý0 ·þ½û’º“Ÿž?=>¥|¦j;»Äá¬Âù0:1w¨¶öÕ$æ>ãüþ„CDFlÍË'Kò>]}%VÓÃÚ7½G9q_Œòua¥)˜q‚8_Ø — 2/Þ€*:—¤?€ “øæÕnÅ}5GÞwËYOñ$²¡w©½"SN³ìøî ôgt Z²šÈå:šN)s©"Š0—KâÙ%Øf”÷ <* Y–ITnAµn^’6ÕÈ,êÑóf™ŠÍ5æqêõþI× »|-ºW¦8{#Yæt(s{þ­aM·ØãŒivZ<8+Õ]ñ$qMÒðú•ˆ‡îeüoFî¿øƒƒ–è|÷k`áI^™Uî z»„§Óœæ¶ÞªŠ™#þãVQãµÏÁæ>ö­YºX{Á˜oLÈe§d0¢t÷îù1À2ÚÖ)äž]ž”K´K¯BWêÜ>,ǵ}o þ¹ž…fÖçÔ½tñ£ÍóqõC9Ad*ðý š~ƒ‚ùzò)Åß.ú§GñshèaÀ”odLTœ-O7à©oF/Q¹¹Wq,cŠ>*BV¿#¨è“xWžŒ‡bXL`ÆLÉöË [½¥líVJP µ8íÖ²1XÁ¯qD‚ý…ûò`üæû1ñ¿»HZºágé>·!<§rsPeî6„¬Üà ©–©Ù-Ý´\Þ2˜´vÉù´|F<–3}ë!_ÌΙ% äf ›h88a)Å.þÕ³Ç2Q¡ÑwJ,×AYÝ;âþCã;ß ï—åµwµ®sY:—ž $7ÝÀÜön¥CÿæuqzƒqObí AE—gO(Çíé†×H-„kÌæ¾±8t“¶WOÑ<…ñÁ°êr¦4fÝwSÜ6‚} æÙ6L$&•ŸÏ÷"ƒw+› ë»AìPê(Ö@w* 7{¨ŽÐð‹x¥bŸÞNì‘h_¼UZnQ³Nœ`™X9Œ™Ò×Xï õâÖíJ$©€©¾lŽõÙáNŠ?é³+KËO[{ˆÙÈÀ²«fûž|$vh"ØžÑx› èPÃà§C: àUb$?Œ´Œ,ZûÿÔêJPH°ç@ií I“.zFªº³Â >åy]2½¤w¦µ3¡i÷;”WÍ`ÈË…}9ñeµøaœ|ÒìKØ@] D 8¡¬=©;qN}ýhwÀÂÄkï¾.F:vl+À‘ÓF½¨!èàmNÒÛHI'ËåB•øŒH&8aÁ"ëTJ¾Ë£ªÅ¹~!œá¶åοÔDGÿÁfWÞwÁåzÐó@ýµ24‰L~8ƒ*)è{—»kžZ¦“!°{„ä>/f-Y Þ1#Æ$yJ,}™c¹›ÜóaoÄ4.ÛÐï\¶p|[cÉßäí­òÉb¼Òž;ýnúwj87;ö Ĝٸ•Ê9fíÈ+8ŽÖÑŸM¯>Dc°çMm-$0!TL… :bZÆAÖ²ðX~á¿NŽwô,#_ éÝ·Å6L@œŒÚCõª‘#˜\qôamê>²ŠûÙ/GµÁ’w·ßîèl¦ð7P“ŒO^|Û;=‚Ç«n/À¢3×{öRÅèYøSûŠós}"Œ˜ÃwsÍø€‘B[ >yø°)ë¿üÊö#µûù{fãR?+¸Îõ¤ÕyÛÓ²—r·HªI1Ôe'€ø–$#ÓPÙÃ^_G ÏÈÕKEÅÁoä…I<Ç ò×éä„qòy‰Bsoæt)¼\Øk yqÄÑÝœÓ.– .üÝX®·€ÄB Ô[/¬ÄàH}#¢žþ!®]ºr•Y—‘ÜL(´y­–/Ëc.‰¼zkµ…[K›vh—…"§KG®ÕYÕovò¼væÊe¯» **ö°OùbÆ×ÆçàaDíù‚y¼Ìß5K–Ó^)üõâeüL¹ h}™x³]hxÖè[dP9깧<'¤—ýíô¾P­’Z;œW"¾—Ųr<ÎL»`´:FkÛ7/úõS2YõnM`·S,^€DÈäÿ•Ëafk ŸÒD~v w~™N;T꡸¦$Xjk'¦¨7²šT~üVÇÀÞhoc¢b¼(–ƒ@Ê{#eí:CYW5û“©²ëK¸.εøŠ¢¶Û~©±.Ø*ÄLRº¶˜\÷i‡ðêrH ¯Z–¸SÃNò ÇLÓ3áTö,gI7˜fb@".ipoËœ ~î1d,½njé&u×àÕÛƒ¢·ÏsRåBvp­4Œþ2ö“Ø"ÖǺlƒÚû‡Œdµp­zñ$cbéJ“W½k>ʼ‘¥éóoôÅ {]E4ïSÅnã"·²0ùظOà¯.ê[_ ­5+ëKÕB>Ûý2Ù»*Æý'Âjm»!TÍ»fM¹JJ¼õñù@>…´eU¿¨ßܳ*J ,1ßKª ¼P™µ2¹MGTi[a?• ÅÓ·®á0._7îï,¤¶W–}ã°¢:‚pŸùrïÚýw8èi süøx ò\}'5ƒ<ª“‘Ôv¹ $†Ò­ ¶MŠ<ÕÓs5X ½Cã–N2«{´½/…ÛÃv‚=EŽðå@rm+.eçÁWPÛ,êÈç|Q­B7²¬DaºÇÓ1S;m°Å+Ôà/ÇlË…õãs%ä¿xœ5eL–刯júÉÇŠÒĸp«h.I4† >Mä«$_¦dî bU"Qýe´HhŠûfJÊOâr³. $x‹ã‘Ym[=àtÀ^¢¨tÒù)•Æ'N¥#z†ÛäˆEˆïæê6`Ì®òØÓ„_¦ÂfY±¥˜õæ,á**qX®ð€?*ÌH>ËÏ…š"Ó‹é,ÉÉVÂ_èµ?¹Y°ÙŸpbË_/‹qÎÞÿš‡ÚººŠ……‡¢å”ÆÁñV–cÜ7]nç«šgmð¦Û§¿ ܨÂëºhÈ(oãÏÉ?¦½·Ï£1±d'à‡Lve|V±¯Â/ÏòmÔ ›!ƒ÷UÝPW{º«2-߶Ò™õdê.Æc‚…'ݲâcÚfÑë+Ó)q¡¾,îÈ­ÔÔT‰Õ»åL0ÇùÆlï¯)‘hC…}I ÃI³üZ+›Àh endstream endobj 1364 0 obj << /Type /FontDescriptor /FontName /OZLTXQ+Sabon-Bold /Flags 4 /FontBBox [-163 -228 1000 968] /Ascent 750 /CapHeight 710 /Descent -222 /ItalicAngle 0 /StemV 93 /XHeight 448 /CharSet (/A/B/C/E/F/G/H/I/K/L/M/P/S/T/U/V/X/a/b/c/d/e/eight/f/five/four/g/h/i/k/l/m/n/o/one/p/period/quoteright/r/s/seven/six/t/three/two/u/v/w/x/y) /FontFile 1363 0 R >> endobj 1365 0 obj << /Length1 2421 /Length2 28567 /Length3 0 /Length 29575 /Filter /FlateDecode >> stream xÚ¬wtdݶn'«bÛ¶;阧bÛ¶m§ƒŽ­Žm›ÛöKÿçûî9ïÜ7.Þ¨5öškîoÍoâÛ»H åi ­õbÖV´Œt \E=}k+ZkK=+lj†”Tبç`jm%¢çäˆÙ™dôìŒ F6..V&#'燣ŠŒ£½žñ‡3;+3€…™•“† dš›:èYX¸€VÖ†@C€¥ž‰©ðn¨§oØ[98ëÙFÖv;kCGS+c€ƒ ð†« ÐHÏh°05ZÙ8X\­¦ökW;Sc‡+…å_áþ¢PtµwZÚÓ-, |ì?Ѐö@;' !ÝGl÷ÿÛÉ×Çý9Ühçà °6úG€äGðv6ÖvzŽÒ³2üÀ2uø{LÖvö4ŒÔ\VÖý?ìþÆhHp´ÿókhjoc¡çúçÒÒÚÐÔÈôãêæÃl`aý‡×{;=+{# ÝÇÒÙÔÁÄÚÑá¯À€.6v@{{€³©ƒÐ  góî¤gñoÒýçÓýiñÇ`¨ç ÷ÿ%匀¯¦¬?îŠÿÉý¨é_ù¶wÔ·754Õ³3þC)þ¹Åò/Šðï€?Ðþô?ÿWRÿòoÉÿ§þì_3ÿGÞnÍׂzÑ›~ôÔŸc>Ž4ZêÙ™ÿÕÿ‰‘ñ#hƒ?,M­`èÿ °¤•‘õŸ™üËnèhó÷=' ýÇì(þ÷D‘ñÃñãó÷+­µ‹;-#+€–‰‰ãCh?˜q²°xþƒ§ãGõ­þ’à^_™~$tÀ,Í[PI ø´¹ZŽÈM#8dj¥+xX-_i̺#Mðê‰Çv[9)˜hYP©J¼¨üd›ÓÑW\('À ‰šÎõæü­(¾ïÖK4ê¶xchœ–‘ǹEDaHèo{ãØw_¢õp”4W»¼ÕÁFFÈä² p¬¢vXÚѸ莥‘ršá i÷w²¨&N½¼õ& =t›£™Ôž¢¥äIч<Âæ&Óµ Â¼¯ BRGÛ·´¸×6õñÖù¨Ã±Žg‚ p ­o¥O UGª{†GÏo'<}liÊC•[qmVØ0Ð]ŽÅ) š=~y ½\ݶ¼û$Ù•;¯h4Õ…kß@:öJœ®<¯š?¶±ÖØ™2„•ÖPÖ®ÝYÎŽžò^»ò7™÷ob‹æág¸_tŠƒ‚a’ê¸QÙ¦Š«œ?- @;бËÓÞª‚! ¾Cœ$•i%æR z» é)O‹Sð¸óÐy–ùœ,x%ìq†“¿p—Ê™ô*ݨ¬Î³f‘E4GÛ2ö%ê×ádâÓ²„X“ð£S^Jsð4³DŒÊÚÊ Ð¿ÒìŽá–'K[Ø1ç4WÚÐj¦BörÚÖr"côÉÞQŽ¡¿VîgK.Ý©/äHÔzíì;½Cw¸òÏKæñÖ9/Kny³Ú¤òÆŠlÙߘö‚öÇU¬ ¸þ¶?B  M²à½ç•„ÀÖÞÚs—¿nåCú¬™J"@×€dq*}c¯°76ÖU<¯A—ÝJG°ñ¡ÕAÁIÞë ¾¼r“ú*p˜Ã¬ºž[ž{©VìñaëÕ8þÂÃÐæ ¾…î›^Bv:h2ó4*þU®[m’ ã@ë³T'‹*w„Š¥*nðˆŸ*b§}”ÎB%ß¡¨.(èä‚w—”±€«}h&øÜ!–¤ðXþ{¯¢d§´èç╱Òüb©ôÀvçGWÂqÅí|ÒGzHú•Ù¿*{ÎöÃ9øzFRçÂÅOB%IC–oÃiÃrÉZ¼:©qòÝ›»Ù# êÛr”âL¦Ý[öLi<·¾4íÉÈâøç °¬¢o>ÝU ™!MM“`6¯Xý/f«[7K`ÝÒC1ž½;HÀ´G¨(n\€™Nö8¡'kgQG5·Õf,!£•1(ð%ºòéî\7–㺱^à¬ð7fQ†ÒJéÙ¾Tß^Qº°è ô¦4Ÿë ÊX¼7°<Ô”ì[îÊÔÙ}UÎõ…`QF/¸^dß8ŸÃISþ¯Ø3·õm=ñ|(»ýçfï„LúuÄ€YO"éËÒý9ÆÖ®²²ž­å~L)óg}ÌAW'N nÎ^LÇo‰ÆkÇÇ1#Âðb½Ds÷º²x !| œ8JÊ r”e6±ó¡ôKÜPf°íHŽÛ½áìQ¥<'q | [·èô“Ǻì•]Q·bþHY ãøõiõQ·(ót%ybNc +*<íá¨i„™ü_ôÎÙ٤ȯ\ŸOs9_xž–ýÛ8^Gh…ã_ÙY‹yì˜J¨|Kƒ‡¤íÛňÝÇÛŠ…XC ;‚û¼2XÏü«OÐ#~Qú^~JeR”Çñ;9âˆF£´aW·>9åî…˜b\“2©©Ã|.RÅaOòS¡¦ô{º“ÏãÞÖtJ*mâÏ(@^vÖ?`ȳ–´ÆØ ½HÓ éU8Ø«cT>œlø#ž˜{ÍÝóìÝÁÏhmÕõ1è~u©äiP÷¼þ`[„õñmO’vC•n9ö®æŽ– -ð®%K “6è~d%“èêúN ]Z$TO§£4*‹³Ä$÷wî5~~C «G :Nž7­©¹n•ÏSu¦Óñš`ø› qx-¤ÑvÅi‰TobãŽêNÉÓã¼IS‘7÷7«/5ãKÞ¤ÝÂbR nð>(«½ãÔˆ!?}¶ôÚ‡\³Y„e–À<4ý֥ߖ©Eu5rûÖKòk ë5%JW¨‡e_¥÷9»•Yàqî< TqóòPýXh¾§‡€é³åV\A¶†™M•²ñçN›Ã5©UŽ…Œª!äf³OÃ7¬Pç»ÆL(ÜMS— ó»úîî'œ¯€`VØh“D‡óˆe^>û %@,©+b’^då÷0C/4×ÉzVDhÎÿN 7@Jÿ)a89¼aµÊ-W–õ%W˜J<]DzôKp¼’­¡à¼bYö;?BX)î£Ûæø7øê1¹t%ÆëV6a3NëþüO¦—©--»pºZöÍ\A9 Þ¼öäp¿ LP¤Œ3$À@ž Ä…LîÆS¸÷ã,_FüH,TÚÇ¡Eû¹Ö Ó\‰»gºg¨·Ê ܨZßìN%ˆ·¾Ð¥ìLO[¾êj u#€¿:Áwh±¿>²à^˜U­:Fª²ï*/¦ñõŸ‰Ån}“‘dÅH…Û¼=>ðâBe(õ¹³–z{`wÞQ3[` ì祗Ÿ—ÆIýšua‹8!"Å+˜?Š +$ø F‡á¬ðÊD*FÆøÚm€}Æsžg Æž<Š{º;Ûæ†ïT!-øDûláo5 ÆGŠKTÞTû¡²s>»×5ùí··óÐô~w¬OêÖ9‚¶¦øåfÁr®w+{¨£¶ë&ož(½TäËá·ÿį3ÈZäè\ã¸$ëìé[÷Bî„™ôþ,2tB¹%*#Œ;þcªa.´“”ü±=šæÌ÷ A×ÛY䙬u¬í»h ‘V›•ÒàïîÜõ±Ý‚ô9×É™†©7ŠŽ i° 1vOwñ˜Èf×°Òq  ƒ³dÚËõöãM°ÔôÆEn½ê´˜ŽÓæš‘ÜöÜ\bïÏ ßè5…+©{”d_è8f«€˜_ÊIBª¥“"ü¤”*/¥ðZ¿>cÅ\Ú¢Üâ”ÿy?$¦ŠéÚõKW_²Âj]Bé8%+ƒîOÛ¹Zu£Pý‘XõKøä-é3AøNcáf*5ª•INzŽX^Nšæâä'ÙXRÿÐ褷ª“æð0ˆSë¦@õ¯/`=;n&<(`;É7ù¡zÂ.”ï¦T¥úe¯ ¦ùÚÉÈihb›yI¶‡"”ÌÇìP˜¯”¨¸;!A†Z…~»øçå—8’ƒXñŸ¢I³§…ŸM¾9GVþê,L p ‡Ï†1 ëð”¸Q& ÿ.µ}>(ðík@TFÿõ·5…MœdÞnšù-ÞÜ AêKÆìKãWö%Ù½VžPñ Y»êšT]Ä(YõpYŒá¢Œ*ê='«Ã¬ÕøioœñC ÓŸñöœ˜‹>Ó÷KÅ×Ïîc·Ä[ kÈŸx†ˆ<¢9СƒøÇr®‘»X§oÜÈÛíÙP_"–šÆYNžÝèqy§&î›Å¿c¶Å–( eå ê@fç`-ÔϾ2DÙ ë¯(F{UCJT:aþBÌëJçç¯xÔ¤Ž½wdeH&†zŠqgk;Zø\¥æÅE¤w{Pç2 PëæB³Ú^+zyòòæCqÝÔÒH!£a\žó´?R¨zD•¬êzõdý5½,®C%´Îˆ õ´«Žœ×¦%J¨ß÷PMU³8–z¨Z¸4v+j˼P»%@„JŠš•_ESœFAY‰×)5ð¢R;´j¢ó9oIdŠ=ÐS€ Âõ[ƒ…W…!„G¶0G¸Wê|‚—Õü@íýž“ýѼ~ÅÒüŠ?Z«ò^£ï`Z¶ ¨:ÿlé–ešÕêõ$6Áøé aF6»ªrPÞaÀŽãMчPÎgNkÞŽš ×ß’9(pz/h-{Bêæ„bèýétzaéK?ö»w£· ¦] žqb¿´*Ç͉¬`9Ù#ηD㺠é°L9ÉôûÎU¥ÀÊWH>¬þ=;[¦OÛC#®Ýž52±"goÐ PGsAÞ Ìm’06êmvÙd”ƒqM–i!?”ÕXIƒ-Â(ˆw¼\*¾º¤õ§Q%…"ãGO™yt2ÝÙþ]°l»‚HÁz«¸+D؉Ó^ù´ÒÅ]ÑP´ý¾‰n+¼Û£O}—Ó}*ãëú©W(&|KÊá©j¡ãY#¦l×ÉŽG‘CÔØ+5vûëRÂþI  AL-ĬÊ7f†$$Ýa â÷ÐJôKëÚ3õÇi]]1èx×ææ1m¿xþê]+3±¢›€'Z¬O(Ö}üSâaö¡6Ñjc䜀#h“–‰VܤH ž·]hŠ¨žcSß ¾Ë>™W¾…d vô´R¨µ—Ûm“|bì)ä0„O™Úá9sߨ‹žnV¬eL‘²ÕK()c’G)„­ÏEvÐvzoÞ?Á¢KúáÍ;ë[:Ý ßRYt[AòÅì‡Ú,&CÂ*Ùz6‚˜u6ÿÑËP;ž¹¶F•¤/˜–;%°- Ô‚rêqzõP’ГeEJt•J%ÉÈ#º~š~r(‡Ô÷Á„ ¤óë;å«­€XNÎóÞÈGŸ;"…(b=•m3¶¬³FyˆwÆ0ŒÏ^ú[¹Ÿè~YSÇ1¡5¤­+¾­÷¿íÝ娧ÝâD·,ìê‘”À —À®ÃñU`jBn…—ÄÀÕ‹Ô`†s1eævÙ„¥U"þØ»ówK`¬ vØã-#›ÆØ|óçG7Ê›4Sܨ)Š+qÃ@ð ¿ ͽÆâ*¢>1Ë7{š_µÿ¥¸ó³˜º°òVZ*×½ªùì½+*-#ÝÙOÒr˺‘–Í·ö³ÏÈí@¯½*„añ6=!lšWÈ+Ö$¤£¤Ÿµ¼KíG|ZKâÒ#D6°î|1 úp»7¬nVÇx&¹tÆ §îŸKø͇¢çÒ…¬–…YƒHðÆ‹9µ ÓFU°0Jçîð–b#ת)¼Ûxë]ŒK'=?7Û8Œºּ׌ùFõ‰?çå2ùzû¹-ÐœsƒºÐmî ßÿVZÝ×ÒŠ‰p,@û‰)yOôþú=`øYa0“l7¢Y˜£s»Ê£¡QFŸ‚HÕÃB]¹Ë ŨÊGt\¼QïÜ0P¼¨ýé¤0× ÛmsɱÖà`çsâ;âWµ+‹õUñY~)„*"LéÀÃœß.T ø€²pT TåAÃòDšÏÄG§‡ÓÈ”DÀ/®7IêÂgwX&3'ûþghÛµÃiémσ˜N%@òë®õëþòÝsT–‡×›¼Ö‚kî·Š¯•þGIö%£r1^Míø”#qp¦jFÉ_ÛT>?„åšÜÒvŽ‘¯áÄ©ß­ç$ ùÖ‰Ä?§·Ú*Ÿ™Æcu‘PÎéãu 2.w7òÔÈO˜ìf[×ϸú‡ÕÓämM%ƒ¿¶™60Úý^±Š% ÌÏN¹BéÊÖì¶å)³‹[ˆ¬œóúñµ` ä·lÎÁÜ£¿Q„bˆ祱|&¾ß†• ’H ›ßMb’Ûé¿K)Ç],°Ð;Ëx"ÒHü!9[3VÜ3ΕG>v‘vŒø]¯yØÐZ|íT ›kÓ‘Tä+M$¾:vâZÒR8¸¹4zyÅUç±!®ü¶Îª•ŒÞq²Fk¤#rVLÖ™®“gáÅŽåK@,ˆÃ°Àøa©/úÚ“#Bõá쑉ÖæWÍA ©r,™ÄŽpDËÚO߶Äøš$óJÞáé]g[¶î/»2äˆÃ¶­¿ãWÌšô÷¦;•BÌ 2´"IÌ5Ò;.Àžï5øP@ÔÞÂ'PI_àJÉT†ñ A›õàQ Õ›z«~éQžA$IJÀb™óz'\œÖ)À}ûLdaÂ~Í™µ–TßEZÈ)Á¥ùd´ñzSØH²ø­’¢Â&Îßo ÎãÔ!˜±¸ktÚ«;ˆ+³¿•ß98¡P­Ìt@EÐ#ƒ^ò‘†‰fzUt"¼ip|µm³jZj…B(Êþ”û’Àm0ïÇvc_é4=Pœ±Æ IÕ¿­{/ª!¬šwAMð»ÓŠ"¶êÄ´wŸÁ.‹‰#^Ñ©©äfuVA%²E=k^ið tÀ|•ß”lfÂ$Hé Ü„wÅ´«î¯Ô¥–üÄæ%×|Å1¡'ú2‡M dÿ{p9F(YoÞ8’P7çB°\<1À;Š¤*'”y6ÉÇþnNÀˆâ9ƒ×ñŒJÒÍÇë(Úv&’µ+•z‹¯ø0î­^:é}ð„PÁi¥P)Æs뢄 ¯Ó«9Œiú€´ñW$3 Ú‡uçéX„·8oÊà¹1ˆ> ®Z¸ž(Üĵ‘ƒZbÍ…² «7–žx ´FŠÁ6êéi ¡/Úð*§h#¬5áÉ×Êý#ئ{c»ÐY²ü@…lý¡n+äê¾ÍèÑôÄ }É»*‹hÔ˜#rÞ /4ˆ¡eá¸KÑÙO®×/"F|ú¤ëÙqÓâñn0P³N(ûµ…»uBü¢õ©Á?8ºŽz/TGeºÍu½„¡hAN ¹ô«å$Þ½ƒúÄèR@1°ØŠÑwÔChh– ß{nR—!B7Œ òܳÏzAo0(§Ös¾ýEg÷†SäecEÍ NGª·×R™V,iª®hB¬WAÔ6iI­öÕÍ"èïn0u½ä7?ó\íÞÀfåfÔRÐïGú]ä×Jζœ_VÉ‚'1ˆo‚tÛ~;çVž‘§»µ×¢Ø+YH>Éœ3ü¦Ô ð%7ì‰U8$*\…sP{2ÿ»h%‡iÒ½…ns°:S ¨0‡M´…uŒú|&l&éºbrð‹“ñª ícFôRþ}‘,Z–pU0ΰ ë\à¶èOJ|°i_Õf’#‹mEjÜÞÒUJüÛ¬õX2HIid|ú¢aö”#ä²(P5¥ÆLµ¨¾„’õÔ$0 ,ÄO9pun ®Jˆò¾_˜–ûiü­ ç·òwùPù›–-e@S !LÞ¦G󆲗ÌúOO-g+FºkÏÏ⃨ö´gð;rt„@–ƒ›µñ!$pd|£Â·2£êí¥£ÈJôdL¨rû"•*›•àI1ó ;güYc­bÍf3ÜCÇNÈõ¼$3ƒf%BY_ꨀÌ9æ«ØgI@Û!{Ü»É<>rB)ˆÞ]JMÆü”¸ 3åÚ½îlá³®C˜>²`8aÞ `,6.÷DZ–U²Gè"¨ˆ ”‡Æd¿|ýƒžõïº;ÿ*\¶D$+×s”Xîþ$û匔R³¤ËÆ×ý†JRžì– Fñ’ÎÕ ï"Úg9Ú+ mTË­ÏïCÙ£á6{ à J‰ñÐ%¿ç²ˆ pºQ'FÞšç^~³b†Ã‘¤Ûˆ<ô´pâ> )‚ú"¨ DÄVmÇŽñàËEc3<gÊÁA.Âóæ‚/¾°s2Ö´ò^2¦˜~áH´ŒRQM;½<ã&Ýú’›Å¢°(Ùµr±ÛðŒÖ0ïzÃåYÍ 5O‡p2#*èß× £Ã¡PÔº‡¿JT à‚¢½Ø|;¥^t#vöEÈFç>’ZU :ôÌ… G†.»éö:˜wµž½ãn4æD®ä÷PÇ9c—†"¢$Ù¼Û¥´ÙJEvôqbÚѱmjuòŒõõ]ˆ¤§¤’9{åG]»'úv/²}|c}¬’æÕl¥P(ÅÃbº>6MbÖ~NçŸEbŸzHT<Ö¿Ålÿn·éÈ@^Œo-ÁaiýÞ`vZô³ñpÞE}Äxm¦ãpúU1ßý.¢~]iÅK1‡ò*ò1^+Â]¾TrÝÅëdüð7Î —ÄKîíéµPL@ANY»'Ç…á÷ÑitÁÇOë_‹„9Kæ¿A Ü~’õ4“Ðc³7›†sg°ìš­Í–èÏikàó$…VÏ~tïtDo†>»Ÿ„^땾þЗ«3ýÁˆ!¡£¨¶¼bï[ $Lt˜{Ö±Wê·V´^Ó˜G:]À†×Aýƒnf“uû:³ ‚AühBÆÐ’ÕCÈî[ZKg†¼¬0…Å£ŒRó/¢£[¹Ìç˜áLx3=><"™æ φÄoÈßr ¦AÍîÑ}B6ÜÑ`*5kµ_béP=Ö)˜¤box1"˜o©ú™\>ùáUDLË hO³ÈÒÄ¥M”ïH>ú_1råšqeXt÷OMJ*@B8ôÀ—|“[ßØ{È lB+Õ­!(7MSPŒÞfP&?$¾é2ý ¶lwÑYŸ²$ª“mtL™­æ%ç@Mw8 èº=ÉÁLè6¤{-5«g+8ñKAãÂʼnÂ˼l4Ç„bW@¾u¡#­W•‘×–NÝzçŽÉEDÎr²~šgv/4^¦0ñ™€hš¬o¬ LÏ—Y€öü{œè‰­zö8®ßo¥Ãþ†jÛð%«‘lwR;PøG$j&%oçTþC}iXÍ4âññ/ÅŸ[§H¸T¥^ûu'0 ½!´¸í2l¡ì”Z+˜¹Îô‰¬9’8e'ÅÐ(Cte5ýnc{7[†íÄ^‹MàÙ\էў+ÀØsÏFÞÙÔ¬»eUÕz3[×602wÑ&Òåƒq(ŸvÒJ/~)6=£| >L⽕[lb¦t¾8 ÃTç®2’'`žìSB½Ð‡»f$]A±X'a“oUt£õ3• ×ökv ;„%Q*9ÇJ¬æƒµóbСNZÞ™_”e·DQ$®,z4àè¾ø÷±Ç^—Õ¨ÈÀDo ¬¡~¥½h0Á‹²"Pkãâ!`rviðÈ%ðJ í©£˜ì±Í”þàt`—™¼Ujš”'®É>UîÙǠòÃùøG`QL~…ë‹H™i…O¥ù¯®ÕŽÍË3$]¨èÔtÙЪNJo±K•¢ô[âØîvÄee[^F ªIDÕ$Q±îhÊS¡jH{er!ûtr!‹ó“I2âMÚ¼=Þ²ìø¯+u„ºÝ“:¿ýï ÑA KÂà¦2i1=U8).Ë°žndzªq÷ú?ZBÊ喙ؤö4ÛP*©‰‘²aÔ59ö~ÓuË¿7f¦ƒÙsC<‹Œ[* ZwtѬŽ¥V’äp(‡ïHV¼Ûq¸ZWvò‚FŒåt=nƒäNç¯^øxå쨼ˎ $ˆrhø½Yó‰®KŠüœpCa"r¿Ù_ùK¸Öô¼7Ú—Yñ;0|¾NbM)õ KÂ0»±Q…œØýWŸæÒ;áP»I-uãT8øýÃeÉçEð F- UWኒ¼~€êýK¢``„&z1ô²õsu«:–„~N†âô8¸Y†Ó­ºÃÜwÿâß!cy²+‰û2žè!çÜw]fC{¿îX7‡4hvÆÉŸ2å†{çë´%"cãÃ2&Oå‚ö Ô–<‰x\¼öÞýNœ4,7} •°%³AÑ ¢÷žšëjœ¥ì ÇG4ÌOiáµ&¦Ÿ‰4:€£Ts^äéüÚ)¦^„¡ü‡íš¨5ËB*ë'9ótÏUú’uÎ?Зõª?–F™ÆŽ ÍØ÷ߨ8k±-QIq¼´ÕƒJ`¨ÒÁjsDåa8¼è±²2M4ûŸ{›Ê¥Æ±$T³çóã ÓëkUµx0úÉhq§„Eˆû ¿¦ÿl·vK¤ß”/½|熛™ê)QÁÃâKs†“ˆ<ê'gK.ùɺŸ˜w1°õ{Ï€†ì—Ïn+ ;D2kœý¢ù/ªÆPÃJ7Ò>?Š žÎnù1yd^ËúÌWSyô4¿Na38ª`“Ë1†/¥Æß– —eÓÌ€öºC5Ç‹)/HªA¥Èr¹Å…ý(Ã"Eœ´Q®xëxƒ_¤IúMÃQD ŽJŽí·ÇÔŠ0jŒf´<‹ešQU ,~yU¸ø¯Jº¼ôöÏ]Éf²I>ßdzS?¿aÅ“Ï°Bbd~íÂä[tÏ\[–YÉæu˜s 2ÁÊ-V\!d|úÄ/hõF¼¤­*ê.G0ñ‡[@SrÞ^@y¬½Tú§?ÍFcý£…~ÿôœ hô!%qßtñó×àIõ 1—¹X,¤¶¼ 3×jâ3S,öª•Ý)ŸÚK˜öÐÛeÑ´±ÓÄV…†z•Ç@õï ïëJB.±Œ?ìè‚ iËà2¨+¦0Þ»QåãP)…~T‰O¨¦RaáìZ“¼ÔÙüÝÑ·r|Lž¾ÞaÐÝÍMòx¬2Ç /D|&Fýl¨hœ&¨3·J '¬’_L‡v´m]8-`«Ï ±['¶n.ဦ_[¡ôì.Áõ§m [Æ:_5Eè´”œ¢Ž>$IèG$/yÙÜB)TyŒ¹©>Œ”‹ø+ž¯æ8B}O,‡cK€ªû­žhÔ ä4¬aKt}Yñ{,Õkl‡LÎåZxùòÎñƒDÑàÈK iÑñeUˆõq"byqMô·RìwOâÇcŠp6Žäm˜÷ŒØÅ¿k‹{7ñ:R\Rð=”¤º™‹¦û« g…Ÿwô7Å}yTüJJ{a Ct~õ–݆»eÛ?ªÏ"Óº}›%RQP`IÀ>mð‚ ä‹õûìdàõªCݵ¸‘LÜ°òöËo¡kå{]Þ•™AÆ]OÐÞ¢§Q„€Ì¨½Æ¦bˆaµòç®PGîhöa—væž¹~£iácçKÔcüÔùÖxó²A˜¡Ós1 $I`ŸEÜ÷ÉWÂrºõ ªXÑ—Wê\œk7b·¤š•)çuLµ$Žù=¦+;;Ns(k5-žˆØá&ÿ“Šjƒ¤­ áh1^Y ±€!N韌üۣƷ6SÔ-ˆ-óó•n6d¾^bžM+mJÝ7ƒÊqƒ¼9•Œ àÆC1N·Ç(üšv“7Ö«”“‘ 8í|„ð-´GV­Cc3Xd…YTиé¡zè.{Oˆ»ú ‰iy´QB+À0dŒ1Ñ£WoN„íjb”[VÇ+fX¹K éez»Ñ™{VµLdS©ópï2¹nòi¼<›£D “¨fÆ:Ë“Z ûÖDfÇhR=ÚwÇ”ùFÏÚ|‚ z°Û{A —pÕ¦‰ñ[?8ëž"oå8"§#mEx“ØL³µÃ,×ý»ZÎÏŒ‘5 oejÂPÞMÛ¬!\N¸…?÷nádÃçA¿)“Ù¸ÂÏŸQ¾•523Ï×Kâx~#ß^8%GÉ8K­žK"šÇò“ÀGc¬ª§¿ÐYTK{ _ûaÛ è{/šb™Ô˜—^™y%yk²R\Kõ¨Cadƒ@¬úpº9Q]*I$ˆ‡7É*e âZÑ7¹”s&mæs_õDíZæ´¨vý‡;ÈõY¦VµùD¦ÃºtmzçWÅOÝìøü™žÝ-©eß–œ廉ž‡¨ø9~û认¤]ßrLÀ{ê" î:éC*Mÿxn»œ¨+TF„ ¥Û:±g´;-ÕÒ1ýÎÇ[yìÙ‹Ùp òî·ðW €Úå¯Lwte-ƒQ¢x‘Ð4àõÉ­‚(8)½…¼ó[ZUu^®wÃÿ> ýéòòó˜÷È£F#ÌWÜpj•Ls&1Ùñ؉…&\ÆtŒ§OYÜ¿þÂÒȵnç"óp_æPU-õuôÖ¬·æÎ!‡Ü´Šs›ÇŸ¸T0ûù³¶CôâjxìÎ"+¸ƒdfbUUzKžnqŸêóýñ×’ ³"íènƒíÏ\ Ôš0£f{ªtÑb¶øqp:AÐé­aGö†ç+«‹ü0‚ŸÏ–‡š™ü¼B;ÃrºžßÆ aª±T/°f»évpò½Ó×ü+w…NOâäƒ6Y'§»&þP#›sá4´ž[YΘaÁ¤G ‰ˆaF³vð+§¿ ’c¿ä \ µXJ²hµQ¿­<«Ë"ŽêSW囎‚qÈWqU¶T^*^åWáÔU{¢óõ½ÕWº÷D©×jU íÛ¯ÀÅ »J=â;ØÞÀäc&õ¯S)MÏôV¸DÎÀÚökq©G;c=YÄ5ÃEsÆI4žÁ3Í©L]¸]m¡Â™^Æå…tºç7¨¢­ðMBYœP†ï>í<-h’KR¥¼dïIùŠŒ§‚*  —¸fkÕ§$Ðçžú¥ßqs|†{·ÿ«u7!å„Ãt“I["B=ñj81•=N1GTrž_63‹™~/WötlÐ|›ÁÖþ¦XöÙëJ¼'¶ˆÅÞº¬è®®È®áœ oaíµºáøÔ7>’U(DçœRœ¢Ë¶C>k/ *ÏÆ*£&"dM¬èÄ)¥àµÎä;åBi_ê1’¹~Œšº¼mv&„Œ£ÇÉ­–³Š˜XU,ìÏMûW"¢×‹”ð÷x~¢1£˜7Íï{OñÂ]"EdŸ¤Ý:UÛëh‹/a8ñE–ZìâP~ñó¾‰yYOÕSâÜÅ Xè1Ñ cê´§¹½ÐFYóÌ) Æù “ç¿ç,‘™÷ÞÿÒ.Ø sJB¿î¯âO[ŒæžòÝ!î»^ rœ†¼ Y†Õ…4ÒAL‡sÖˆó¦;ÕÑ•Ù%ûa>¬ÓÏE\>ÉêÒŸ I¿ O©1F„H9?;O™éM@ÐÖõåù³«Œo”5¿5;µ­ç>MÍ" ‰ÉaiúK!±Þ:BÕÊØ.åžl„éLhÔ¾¿ëµáˆ ÝøæM¿îHó¥[Éäb~nߥà}3½C-RÐËüLcZea´« Þ¢N15t#nýª)S¾Ák•+ç®—÷U°±Mjheì=Du˜QÁY^YM¨•†O0|&¾örؤoïä²Íž¿©î%%Ö*yâ·âc©f$&§š(|¤üT‡X®ÇôÈé±=¶ƒÖÙ©3Ê]½¶.;f³fVrUi>$±¼wAåùÎM¤L+-#uæ"útKJlEl–8ŠxOêä´‚…‰/"3Ÿ¶@Æ%ûŠ’¢ØBÞN]-%½+’!{ÍÒf¾Þ7Q Ç»ÐûÅ­³²‹yµ¤Š®d ~G§"v·S„'ŒÝ·4‰uFxødYÄ1ÙW½ƒ/„ÕÕß…~ÔÅŸß+žKÐ_Pÿ”Ž5ÊHy¨’;Ú£wäàEiíek›ä­ûÙSÿ-8<@‰7 Ž8­)J&™7-²3¢Ä+6áð o5eË'¯§ieóÐý#^”uW~|Ð R?]‰ÿ=õkW+¹÷Žn=Ú_}”\Wt:óËìxƒÅíHé_£±Þˆ’:‚3D¦¦ß„¼4PÝñrͬÜg¬†“räS É5C{_ã¢N‚ Há¤Êªqë-ƒ?b,»´aúnjªÙP€õr*’²SgÒ‹c3£ÙfKq,K߉Ñ÷#)iÍSð¹Z3ÔëÚ‰Û'¼+ØŸ˜Ekû‰,´öŸ:m 7VƒÉS1ÐðýÅ&«Øì±?uýFvš•É%¼!(¸ïð…Ëï?OÌ ;U#B³ë¥BÛJðú¦ãåuÀA[ìï-°q´:R”_íè'®ÿñEdÕzñøŽC¶®X»òí®ç‹^^OÛ!ÿiH*~b=j’€†ôq9h ?éônr$øáX°ŠLx[ÏZWí»Àü¢?tžAÐnökÈï’z3Th:Z°L[cæÏÒª½ÀR"®ØcŠê|ñÂuñ¾wr}ª‹½è´U"mnê`ȚΦˆ· ìZ¹l™ÄƬEÚ#:0ˆÙ\VmùhÄÞWW°–±gz.–òOG~Á¦äŒ®¼Ý?â'77_˜ü +ÜÊ@,´4‡ã4ØÏvB^´E@ó¦:Bï*¿ƒä…±‡˜=žŸÆ<¿iÖQÙÊÙ{»gG#À~ÑpÍ=¨Há º:w¼®jqç¬9KпJ|&"ù|»ÅfŒBéãrªüéRîno‡ #U²\tð›…hÜé>6%ùÅj¼+σ[ž!g`ˆðÂaŠf™)$ R¨YW„„é¬fðÜ9^Ô¢|Ùì鋈ÓÿLù&ó½ÿ,@í-ï­N]³rËÆgõúÞ ûcÃEÅ„=x#M%×çÐï³2j‰T™ïì"„—{°ÀÂ¥ßf§y¦:Q7õ¹W±ÇɦÓïaH(ˆÛœõB^Èg•8<ÿ*Xápãäb©‰Oª%¬Ñð ±<9X—ï%CÎEÕÄò ¬ëÜöFJ¨Œß9å å‚âƒÈ'ªNæ}ŠA>‹ï 1´]^læê8C¬@¹àÂÏv-g¿(Y…f„æ!¯û†Qk©T€É3; Þ+JUD©£2†ð^'}5dÖÊÐ ‹–? ÐP•ëæWs¬‘ížîë¦q· kÏgÄè‡äO¿°3ΕpúúEâ3=ß¾ä‰qîç´ŸEi/gå5P˜ÛÓZZï 2ÈÎÓí¤‹e ø]¥Õ]½#A)Æeha¡Ì;„šÔ;öñÁÒéßG ͸‡Pª/{ùA)C7 6¹©úMÌ™Gµ|‡¿¨’Ðc@@Sf™’.€×Ød"w½ yéø÷ σD€íé`úá¥SÞ} dåú†;š_.òPF¦§æ£·gÞ]üzXUó€ÆÞoržW´x2›ÙRØB‹då-GÂNV´Ä¹®íÊÞ’¾Ìƒ$Aê/û©]‘o"ŸAE ÇUö|ͬݿ^ Ó.×G3yn’(é„~èò§Éäa×8VIx¨ôÆëgmi¯È—|Ð,CJÖÀÕÁGèLÉM]"ÑÚ¬ìˆ`ð:XšÌÔ€îº]="_Xáÿç/´WXWO×î–S ?A¿JŽØ8là>™¼ì‚HE‰,®¹U%ü¶ƒ¾AǧÐeŒúª°—cq>þÝ…ø·Ê°õäå¡|ÅÁ­œh*½‰Èo”Šµ‰ºûâ O|íÅ>¥ ÌŒ±æÙ†HiÀN·ÝNðÐê÷„y{waņ¶fsŽ#¡ ‰ªH™^HƒH¦z¸®K¿Arݳþ^§ˆãÕÞ„šªä:+·ïxåèIµ,‚†`4OûQdŸ½Dð¥,\zýéU^ð§`Ÿ8Å9< °k/½´SáØ»¯QqðžoÜÊét¢o0~L8ÆÕkØX¹Ç~z‰Rÿt²È~I®ÕÓKh‰AƒWâ~Ãe*rŒO^n#ÉÖ ,h4dCÒÓ5Lñôa/;\Êçbq¾‰ÕD¾ÈžÊ…×|õ)>áÒ¬Õ)_ÞD«¸#[ið(°®ô'jÖoZ»ˆwºT"‡ÉKF‹®1q ƒÒMçÜ«%‚˜¥çЂn0wÃáŒ$%ç¨(S,§úŠA`û%O°FÿÕˆò½8ìó~¨ÏŒjáv`ƒ%™TU¨Iõ&‰e@¨PlgB¼âÛMH»0Èú4ˆÛ“DàK8>J£Ç9A¬#®¥æÙ¿ø‹wæm«[Ž=åoó]0Økï‘øµ}ÃìuSðg¢„†DŸG£zbšýY†}ª÷ý ('“ç>ÆÕ¶iæÈ_÷Xüöî¾JŽ3*·èEª!#è•IàÃZ¯2• ú« sxB‹1‡ôAÑ3òÑ éä6øò vZ’FΓȄßæ«ÚÒz£X\E¤lÝ l·o¸{Üê[ÆþžÒ@ûoÉQûl‹>C«†éZé9kZ+cD+AltVÔ ë ešãš’ý§Â9 æ~ŽŸ.›÷YÑ/¶:¬?;yŠp„Dãˆo^ÐÝu8šÆ±Ï`[¿ÖE¤äáò2˜`/¦¬Wšúq¡¢æ….XþUû4¿²}ôº$«¢ÑS‘WçLŽï À[RÑiÌFh‰‰êbÈGÒ4ó Ôû·’/_þí4wuň±ò2L¶9 TV×­›È/t[ºÙ[ý#ˆSÂI×d-kÂ=wÅh'Ÿ×~¸ðæ|èc´”ÒÌ0‰LH1¦ Œœïz)üPæ7¿N«@¢˜é¨1¢@ªòp2ûîgØX­¢õ€èÎÚ]òD¤š£¶½{¥øñŽU}.ß¾Ÿ1xk°”ó(×ø dXGíÎý¢?›÷¦n ïÙh„òe;WÆƒË Û·¶ºmËÝYØxÊš—Öº¾´‡4;#ëb ZW²bîoUGcìŸA¯{Õš¤—؇Lд l@2‘ pLÈhH[€4¤fÌÀ·ÖU¦¨¯îD¸µÞ ŠÎ)'ÆúK®ÆÓ!®ú¯alWÂÈì=ÛR¼ß³œ¨ò²¥Á¤NŠØîw³å‚¥²Ä­z/7UuN&†ÜDqU7ôÙ›ã¿M#”D* wd9Ñ›–`#ä)™uH¤S:öWòËØMëüc˜¢[À:ØHìF=7ˆG¡À;né&îÙb†~wk8¯}R5[9h\Ãý&ôüð ÜOY– !¨’b]xÖɧu¨HŠ»u>k[î½þpÍH/°­Z\Ò^†ç=³žHƒÜþdlíã•áŒ1ÇJ—&6Ô’r¹ïGi–Sòpìß»·`ÁëKïÎ¥àP“¶æ—È{Ö³S1J ù;J®Fÿ+2æÍoQ!ZT d$»<Ž˜OôÄ-uëqäYFÏ‹¼À4¾ÏÀ§óŠX&ô ž'ؼ{5EüY+u"Žp~àXº,>û9“&çüLÕlÎ͛ȭó1Ö4Ìîq¬±<ÌðšÇóè )ˆ[Ø™<ÙItúÃAhÇØ6y W&d‚dMžV§söFúq|jœùùm‡à6¦®Î¦SG~­Ásóÿ¿Y d@Ťâ¯oK½YúÅ/”RŠ.]îñ;ÊÈðREÄ¥‹¦ÈÄ°‘ö/v¼ ÙÓ ˜¾#w[xسq£™šŠöU°uõ¢]ô—Íy‚&(¶­€nm3û°HÕ£ 'œö´´|}ÚUJ®Ú0œ©ldÀvKÙË웺‚t KÜ|ä&ÔÊ’üp‹/­—w0[:ªTßZ‰À«:7²xnböc’{Ïì^vX¡F¢Û"@”™(d§€Vý!{cã.Ÿn«Ô–Xþ ·ÄK‰ÖO”•8Ü×g´à C8AèIêK7·¡EÓèÅæÐ(ûs…òâyU“dš*”õøã )hé¤ÿR#žÅÖ}—–";¼kÚzGRxå}á_“z³úû‹]Í¢/ <s ´šm;Or£u3™›Õû®¬—R eMqÄÌÏâàûâî‡TÐT]kį+öãjT`–L—OÛO='H” °S»Ù«‰›²ùÜ_¹q9-Kƒ{2kÊÔLßB«ý2þ¥ÃötÎcã } ¤J„ádè1É’Ñk»³qÕq½l€p;e’h"ÈéJ0mho¸vÄÊí÷Bzt$o€ ˜Q5qcË(*›2ÚÃE–óÒ4J»o½q0·”†AL’¿O”IuqDØNä…ÿ˜_Ì¢õ÷ë$O¶—a&¡™ z6^w~/ÿo‹ÐÝÚ$ù¡Ô õ¢ÛIä÷Ò“lš Æ_@€>›¾ÔÃ6-¡˜#ìtã*)ŠÊ/ÄaC"œpäÓ‡«ë¡rÖÒ„r`«Dn]h÷ûW¤¹Í-‚‚È#u´{ßÓÞÉh999F÷”ÆARvTÑÉååÒsP¬ùáù—Et)§ó@{…¥ï‚7Ês!ùö±j ÍT¿µMsÅI­fÓì“0K?‡ÆÀ÷åþöÊþ0V:Ø™uC…àK8{µœ¤Êlvƒ×ä(u»D 7,éR˜ÞñãH`4¼§tEWœ;7Î>ðѾ)“ˆÿ: «Æ¾0Ñø‚¦v‚ _ävkñ¿õ•Ïçv<ÁZºÿ‰VXeBJb8”åø@(æ‚Á§3`¶Á ¶gØã¥* œ{!5ž'êg9RÜ×¼ŠALu®HPÆýÖ¹ âG‚HÏ!˜măÌ=FŒAæ=hÎÉ2û pÙÖvhBT½(×'l÷ET šc°ôZ/¿s xuIïÜé1ÈCe‚ µ‰¬ÐþAcX*£|ÖhÞy ¶HGäÐFg]£/£bã(JRóh‘•œç•q® 9ùaL”H»P™¬gSïZœÌ¹™Ó}"£™¯_dI“ Ø°Íè+ ÙRÐAµòÜà˜é ¬qÕŽ så‡Í§à?ƒ…`„%œ4›e ÕÇ4Ç[-„˜K#eÇ®g~ûë°ßo± `ª°^çMŸµ\É}Œp^uÝ*K‘nÂXu%]¢.Ÿ©øút^<>’ÚQGŽî ͽ6uâ²(Ë™K·ûµ(Å¿‚bƒ/;'ÉC&A)Ÿg`zFÓÀ†p¢Ÿm ‘ÚÝ_boTFQóæÅK¥­/Ü}‰8´(¢ö½>Ÿ2¢³2.üa í@ZZØQ•{¢!]ÉØñÚðÞ‚6é¾ÎÁM›'.)Œþ SÐ@»R 戦白 eú&SHÐF6ö^ >~­48oë^_²̹D»zó.VÆÞ3iãY)£`ý‰]ÈÏfœÃ¡[ë×âôäÞ›, MÉ5É˼àå¾mèx2=ºìIæÇKÓFm¯¶Ä¼$F±œfeCÕo€ŸÖûÆH@‡ÿx ©ÉÄsë2½ Á°iöÙ3ù[dßjoŒ×¾pqË×øMÀä #Æë¯ã¨ÿÝe’aÂì†k‚ñ\òÿ—çÐt…+³͘Ýhø¾'“˜Ûx|2ÂrArã0Ã}u¾Aoꊉö‘ÃL¦X´N£&ú,/ì÷Þ›šˆ¦ål¢—¢ÕÊðÍXѾÄè ½eÅÐhˆ±ép;Ç6iÕBQ¶xÖº•‰‡¦¹pFß\Ñë€è0=¯ÁÇûru0ªÍðq>QõÄÎè×C§HxdÌÒXf>]#S±¨îõ˜}›©$/§Ã¿êp‰m¿€Y定yc™ìÒáàŠf&Ú@´^K²o°•ÅGÿOyÌíŸdäÜH_#A%mç‘Æ€{hZdÏÛEãMPJƒÝG²É};¿Ó·]RœÚ¿[<æïÊÁ‘zƒfžê6Åf–»“#Ȅ똳i2î”* ±pôujyÓ°>Ä:ȨMаR×;-UJã ÔŸ*;ç± pÄyW)â¼üáïâ? efÇ·’}™I;P Uº_KÝ<ˆÃªÊ㼸šíÕÞ°2óæÑ'£ñƒÕ0yn}ﬣÐ;’¼(¤·Ú>UPÊUí¸+Ûöb÷º¹ž£Š¸¯èé7²~ƒØe™Oå€MºY)Cû ]`eÅ(a,gøŸœôÂi 1 ëÜ™QîÄýɲM²áögü(7¦WŸz¡uU¬¿xæÉßûy‘„zI6MÌ“‚£T¿3‡œ¹þïÌÿ¥ÈVÛñ£³¨ÅÁ9¸¯A:ü@Z­X†ýVź ø¾Â•„²–`w¸™²¢qZVê ) M’e!ñfñ¿ýJ=zŒ«EÄÞ˜£GöåE÷Ù±hL’˜ `67)¦=A Éi(œÉÞÊhà¡/Àlv”/O§:ºJ$}Fðy¢‹ãñ¬)r/ÍlÃþ*âÓF+c$›Ä9ª·Ð‚|ÉD}©0©iùØ §ËGúœqB¼°6è Ž£|áS°V±ÑRjé·ìé° á£+âàÀ‡»ö„#ØÁB?Õd<%‚ HUsjý òʼ¶êt;Eü^:ÖSû~tåÖ«ù©‚DBÆÆk÷»$âväKÇéO…ïv‹;^ãÀ= ¯~§µñ oÏ•vŽ}ô¸Æapº‚oý%ã¹­ô_i?3‚c‚^Ë´Ð'ߌ¥Ýs[‹&Þ×ÁÿÞ'L÷‡Êj¡ÒŒY aÍŽVõ{àÞÖÏU¡BÉàá›Ëþ?"+×û|®¡ZHbÈ ÚÌ3-#Ý<ûИáWÛÀì⸒èÏD3u–}9bž`G„XQ¥SÕiWh®•ïªp¤Ú'+Ç•îï:t»’•w°äd‹M1v¡Â²/}wàk@KMnÕ&o!Ë1_Š–Å´‰Ãƒ‚;7w¤žo¹¨J¸Ï˯uùCüóiôÚE¾ ÄOr™ˆ‘´Ê³­øÃÀé\Öè6þ¿†û•sá¨X쪫ÿâZ|»Ä„¨ûÛaP™­§Ó¤FÚû™î—™ÉàµÌÂ蹶D€­‚›­ÀQ¾ÏÞ©ì¾4¯Úõ{ íØ¥lÒ¯˜Ñ¦,Úç'‡Üôè·Â³k“2 ªA—«©w§JNWjJ¨öÊfаήNù’ëKŽº€G’Z8XEŸhB’•møŽ/&°ÕýYŸ%þ4è¢eV1çìߦúLy :ªWFÕ´Vü Dð‡z‡7m­š –ñJD§_íD¬Þ‡òzÿ³>F{s¯b 8žœ9¨b'õ²ïΕ­Pu•^{§¡¹¸™_¡Ê8Hþ‡gn*GQìëßᘱYÜñ=&¤AËc÷õ¢ý=…(¢ˆ¥J²?}M˜»ÚŒL×7-™<<íÎƱфnGp¹)o&8°näšCþ¥Ãùð‚BÑ]šÅ熼—¨'?c^6ô§^;UÇ'Bðÿdóc:6–³ ‚xg3…G}Â®é £2R é6o<ýóï…F|VÏ; 'oƒPó•°øTHV‡¶>;Bf¨®´w¡ý»äõ¹e×O.¼ªè“Ð*̢ŋ±Ù;±^Aå°á»²å©ÇßµÊ*˹ŠÊÊÌ=gðÕË/7¦©- N„§Þ·>nøáC˶_)Q¦Ðš9w x‰å8|ÉìdÒasj®fâlvíûÆW‡EÒýáñl)°ïõ›Í‚£RF”lwo¥ïSv"®n=û1 ã(Œ"üCe ×únŠA_Š»Þé²ç¾ÔÝáZ·™©l˜§dªŸR@‚Y5·¯C™?‡ˆ‰"‰s^,vsO O¶}¯J¹3…©‡ ™Y͹ K=ã¼>¢ ÜÖ“2@zRÇ-íë£éÞÍS¹LÑ._‹Š FF®†ëµgÇË,Nu¼Ø½?¹šÑ·»n¦“Á/J†ä&{ƒcCf¢ì¼ºŸõ­øtìGðò+PÀÍÒS˜ôx]bw×îU'{÷ò1LÕ™FÑ÷]0ú%>¦+úY r°ý­5n³90±V‹ŸL‘è4Üy‹ðzP‰ÏÁ'Þ#Ú³~tóÛÛK;¹Æ¤«£ûåï¼Ô¦‹:8K<ëû¨ªNöàß;´ôæC´Iú¡""×®Ý/gÀÓTÆÂ*¹8*„©ÒLa3q9w/eæ‘ÞI>z•ˆÝEõØÙò¶,¶§~ìm{øõ¯±7ÒÇ#ÄÍWQO!Ðé}í€L\n‚½r’£`ë"CoÙ ‰‡`:• ˜híÍØdÝèv K|Ÿ†øŒGjÓaâ‡%Þ»}á÷í62©öÓ €eÖ܇:CÎÛÊós¡\m¦x›áîèú}ÜN ÚÍYBe—¤{c²cvÙšiô÷»KZׄ+}V¹‰hû#<¯ ä/•6h-‘L"V;ÂTxñ®ºyàj"‡ø’ùfÀŸ9eR?îO­äûÙ*ûðnƙڪË'ζÏ™ÇE7DÇÃ=ú÷C–ÌmÄX²Ã'Fpýº’™iýÊÊ™Aæ㙲í¯ž6Ì™D™RA§ë9L™jçBÁ‚q…˜Ü^&¡õ¬p»›yüf‹Yi^·„™ñùT}>­{¡[óhÇ=OJ lÏ5hšËÁÓE5;ª¦ØsdPˆ$Q‰Ôö9G¨÷ñ³8QBY ŒL”±î‡tÖ[Óê;ÆÛ 4œ³"Ú>±Ì}seãoò貟’;ÀšCÏm‰<°¢wÔhý@K¬è$ÁeìYW¶­“š·½*½X`‹€Õˆ‘D!jÎÆ~ñåz [*ÑäÎóôSö›ˆ[×&~/Þ ÍóRÊ@½ØÙÄíz­³€Çn ¾«¡¾*Uóp‚s_±YŒ ]çô‰(¦°™‡þetïk±s ‘¾¿£ÔW¯<²§¶=ãð uV:¨®l l5¯ÁÀIãt2uLbDˆƒõÖÈ{4r xçx&´ íÂî¬?~@Òc®bÅH0ç}dPÔ^$ÑöÝ€‰Œ‰è9,¸ÁÙ#í$® š„R!G\Må¥E»29[7ó?‘ â*ÍVÝïBPáñÊ$FG°Z†Šë´T…Ì*4ijc´GÂCõrÛüæÎÁéÿô¢©fªö£7¬U{uºe(xÇ”¤TlLÖöâR([†¸ÑµŽ„ÊQµ‘ÐÖ˜tfÏ‘S Ú©éµA'|Ïǹ§vÏ°<'[ˆ®Ý¯­3èkQ×7$ˆ;w8dÀ÷ÂÙB¥… uçÁ$ÿý±¹O=ïs˜O5⤈oÌýw®ýOp–«Øó;Þs¯ðÎòˆ‰I|gÿ(‚o®¶^wçˆûÚèiGÜV¹ùj\C9½ÅJ5Ö½ª±ÓN‹e°2Ì{ǬÐ÷ržÑE_i…mÚ‰šŒI‚ä ­ä ËL½»Ó )â„ }Ç|pšøŒrYå(àøá tcô÷Ûºó]&˜£5t17;Gñ0|½ê·ÆÂ&Ló»S"ŒÓìjÊë½y¬^qv¾ûš.r¨ VNFO³è—“­BIõm’¯Ààjd¬Ieˆ°ïïMÁaê=v“¡(ÂHXǺUž˜ÁùÜs`÷!æhy*eÝ&vü<6‚â+=Ì8O“ñà1β¹‚¼™ƒr àGr[ðÀ7Þ—õG WÙRÅä>DÿÓúÜ‚¶¬p¬ç³I9¤ü]Âÿ¡[ýŽŸüHHÏ…ªiäû GÈÌa,²úôR×àt‘ðá‰íÖÑÞæGWóôÔ)ÖeÇS·e3úJµ0›þvR(Y‡Ús—´Íòm~·ƒ£@^Dzç ìåÀÙû3O(Í5ºÇNáX AÒO GÊ.k`ÿ‹åtcär5Ë$S>"u´ úˆ­=¹W¯Ýô”– cC;äEàǨ)Nógô“ÀMÉôÈ­͉1ÿ&Ú}±¬ŒÎ^ mëz+ÿ ù•9C?r\>ÚWKèÜp5ØcRB[@ EóÞ‚Ý ñExÛo­3Åz©73â]ü£˜Æ‹Ëµ‹HI/ž”HLA…ëþæ춊„qêû›|Ô.\æ½i옱v\þÅX”3JM;pDÆ‘RB€¤~tjÄ„ y°ö¹`1žõ:&ÍÔagjÏëøõCVì‰NÏ}‰)&Š~ó ‘Ö †Ý3¤ tû&÷áë—in0{Ê™Z»G¯‡ð2–+JY" ž;—åžÝ_9ttÉR…îæɧÎR2į5yØ?]2`H.ÅïF >}ô|ΉDD_ÀNßM˜üD‡GÙ“’Ÿè2Oí-¾ÔgzmÔûƒº³.ŸÊ¤™Ž¦öØ€Á‘åΙ¢J‚ @‰ÑH%Ž;闠‰Î]Ýzsõátçò&ãj7ú*ëe°¯‡bC½gyœ/Y)ü’ÉpbëX*"±·D—FÞîzÖöÚAuµÎÒ]ÿ¿K¯Î zKä|þ|ª¥ -Åü;K£„“`Ëì S€MQQÁÊ ?zâÝr âšL kwMŽ?>“6àt´:¥Ì |5x!e’ZFß#âQžH¥Š…áyö‚ÔyJM #(3÷Ó‡ÿVéÚкÅÇç5£ª«DçQúW±TâÇ£ØR‰ƒ‡DÔ=¢Xõ÷‚a¡‡µkÔJµ RÊïÇ—ôV/y'˜é³dWí¹¯‚hƒ°¹bõ–êâ<šËß‹F¬ƒaw¶|ý&:ql]¹ô0, nÞv:ˆz4¿`oDôÊ5çºsÚœ¹:G: ºlýhz¾ Z±T°I;†ÝS\™®Ãd:O¾´…DžÎv‡ª«AŒ‡¡z CõµÇ-í×ã’ÿ±Œs«ì y§êæõÀnÒËším¢Ë}¤Y¶v+T²^ñú gŽtíˆ[øsäíãNµÇ|0k+ñåµÈƒcÞadÅK •ÕÔw\XyÖ4ój¨¨8³Sö|Ô¼9¼»˜ž÷È*Š@wÉW‘ËÄó!Ru ’€öECÅÜ,ª7ÌþŒÀKwó¸A’!šœŸl&ÜjÕ¾xñþ¯‘Ÿ$ïCÜU¦*®†ÏÓèç9g^éYO&ª³úH+RO«gk=ÔääYú×R™ª[˜P{kd4¥#¡$ªLÉͧ»Ðò°ÚÃ[Pÿm½Õ3Š|{¹Ã>õô±/úæºL¿ík˜l:ß®D‚Ÿy¨Rî¨æw•šµ[£:Q“†žžöéÎkKžu¹ yö-¥¯×›Ð+|ª\Œdó Æ(Zø+t¡|K?>Nú÷!놯¦˜1µ¿ìûèLáv “¿N2í²Ê‘@uZ*¯ BFèÀ’`š¥X@>lƒÔ»“y·t$ð°ÄFO¸·ê~}¨âñ#(*”†ï0D}Ûz›Œ9ÆÒkn½8ZÛ³™º_ç˜2DÜT+´Î ñõE¼· .Ñt ÔWE{^Tgoì¸Àß>B÷Žt_¨SèØ< C»^ön‚äâq¶ +r¢.醱-A|âG7Ú‘ü°zÝ”¸bïÓÛc¿µ²É‘ãÑîR³6RPÞþÚʉûÎïž1{òÃ×ÌѸe4‹Ëk 9K1G6?]|’G¨cžæÚÉ,IDí Õ(ù?TÉ$¨‡?ZyÂÚ4ü#¯CVww°·¿Òxò5VÆ{&=¤8—Ä¢¯<",ñ–@! m¯ƒt3‘"À%Ø£Åù×£«„˜; [OÕ-.êô¥fè3ˆÎ¸³ t¾YUÙJÁ÷/Ó@”Qx·ÄgD|jI8 _5PŸ]©oÄb«X&C¥Uˆ{Ú‹ÓŸ•.]Å Iã:B§›_>€µÜ&ηÜÁ:P!ÔÈÍîx¾8ø:×›–1ŸÖ— 6Úˆéï~O=MV9.¤+và(6ÿÄů¸A©7J/4/š”RÇžã6.†°¡¼¶¼Ìv¹–ääê7¹£? ð¨sZ™¤æ·æ)U ])èÈÕÝÝkáÆyžÑx:7ëzË¥4ÚêH2‰QyÓ­Iæ`züÔu%ø,ËÉsßþº³û›â$×.Üõ²Þù®ˆ±¸$¤l•N|Îj$õ]hQZÜ^‹;xu-æ4vþfO8XgÆ_Ùä‘Ï×±f^ÊË^Ó‰±ú#ÉþvªMÚûtëRœy%ámnp×ôpGaHÄÿOˆ?´ùMlלëb®c¡wKJ°`˜”ƒUHÚÀÈç+ÿãð¦ËúÿÃfFýMI¼'ð;`´^Ž§¸¶ÕlÅ}Ùtí]¶Ÿ‹@«ÛndaíD Ÿ¤›®88FC¤5C~ëì»wcùž`ùv¥_Ǹ$÷ÍÚR´8ýœwK[ctá0Ð.ÈVþP7§úSCö¤¼méôÌè§NFX7Ÿ Òº$aå4Ca³ï§£ã§˜{J­Nnšneñ¨+WF ™ÊüõVC[X•üÞŒ!\– äØ®ôaÐåXŒâº"ê9bˆ„,–)XçXD*v^Ð"£ííSHD©»N^(hRnsÆó#í=‚´Š*JüU•4ïCv§´h+  r×ÎËçCš†!­¬š5º ¡~²ÓiÕL8¢ÿŠ#áüb@Vj8áÉ ÑJ ýšgïCgòu÷]óä³û±¿&5/é{߸gp+ºJÃéO oˆë­í¬l^-ñˆ­Æ#aZÚ_d67TI‡PÕ̆†¿b4`•Ê#ÏËÛ`¿±iâÔüjP.ÈŒõ…´¨w HËÎ#`AÕA/»Ïo¹2“‘©ˆÃq%yæbÚçPÚªRøãË89þpÓ/-lM8f¨0øì´@ äÆAº€¥ÿôn©Ê|iƒ¢¡Í2½ÆÕÒ Ù͹Úyµ§_ï&P?LÈœË*Á%óEƒÁJW@U_@©4E×A±•m3®I@AL?§z=ÈñÓ0)àÝÈ0‹¸Ëu*:tnÀÍÕkÍ´€ƒÉ¾SýÔèd›pÿÓJÃÔ46òÆû÷Ó½¨”ayáw¡ò10{L¬ÎÅWMoÑ„?Âų+Án+¡ÄK kã»0èÑ…²¶­”ÿùñ–…×já:étw¨¤ªÍÁ ºzŸQ'þYëbg=ØLJ-GÉÓé1j0~9™ˆ)ÚeÙ¡"9d•Ó$â՜ևµócOõ¬Â+Åü­(/–¿Ål–thwZr™w¡yw¢%0Ž¼–Td{1¢‹Ç׿\Dbq¾.ý3tÄó8Y%v«Ùó²ÂxfÖn2oYŸ—zàÙ ÎIœ‚T£V«¸Ë\7£QF4ÿ…åœ÷hm\zò‰7‹•pnûUw’͉>°jÕæFÊúð¼ìL oœ}•säQŽí¿¸+owþ¨(ÙS;.¨‚ |¸^ & ×™)›¡1n¾‘ÑÂĪt¾ïÐǹAQ½pån¾œ3 Ø&¡) â6M¾ÅhØ3J4€‚C(Þ5͈q“kFP»&÷Öú2Ú) k‹¯æ0C‘K® ³h×X$ã€+[y¶ÕY£dj.ú^æ z4§k(¿~0èGëDl—Ã…%¸²~Ä_ˆ”«=EçŠ &Æ!›Ÿ`›ë$økïK q¯j›Ž¿Ð&¦2¹]@ Y$""8A”I­ É>1Q^¾(Ι‰Ø…X°Œ˜N=ì%$0BDNÌ13I*7@ &ÉÏÛ/Àé»+ú0S{gòÓX‹D7¡%4ÒÑà~ý° ñoÑS ÇÈ™\ÔÑc66/,“P;¬+YÄ]ÊG»_Á@àºì‚eKlôij[݆¿šóÈIk'!š¢ìTz¯êÖ­¹óÅÏ£$Ù²Ì,Ç„n}: jÐþ2biÁù%ƒÚïÒ>-'˜Œ(„Ìç·pß¿¼¤'m†Òõû%Ü ²$̆ŸAf€èÏ#t„nq²Emü ¡Ñ;ÁˆÜ `ò~Ñ÷i5Ø”w `ušCø:j©ðs­wL¤Ø/Žªºivô‘‰þýo›2š°àaþ&:y¿_´©­¥B 2Ä_•W/¥ª»£%g¦´بE:lÉÿÊ }0*ʳ•í/ç2z '>#ª+0@ñoëþî…•ùˆÞ·LŒb4AEÝì|iV,¶Ò>zƒÜ T]·ÈImš!¾³š™-0ÐñOcÊß"¶'ó »ÿÐL.ú‡×¡§ªèrE¡,K• Q€i¯tÌe)6…ò¶EÞ¶ö¥!™_‡! ËÆßߊ‡‡¤‹¹\HxO Aúú•èæÛ\שÆsXîóôæŽT+ÚúZF›âªmÈ [CÀÙù€ô"{À]ê;„*O4?Ñæ‘À» ×Òšéœ=,¶¼’©Ç{&êKá L&Žš0ÍPÃÆÑ=\pÀâXײWs‹<ÕKk£¬Y]›Ã”õô®ïpÛÒÚDW=Ð&}–­´ ¼þ?’vûI8ýE@è=|3×£e=² #²ÊÞ)jvJ¡é»‰”Q5ϵw˜Ç1€5}!kFTÏùµÍfïmŒˆ¶xU«].1BRԪОÃ0ú«û—XÛ­›ŸÎžº¥ïœWG­„* Ÿ–®4¾ß\G0q4Í„Q5?Iôró‹'öÏn®h´¶ÅIXÉÇÁãž½uÚíÏ7…;î ÷„¦‚q«Ö&½îL<(K½ð†–UfZ9P¬- F,ò]¦4p&ê-e,EÅè“uG²J2LòÛ¤ËÑq‰v;„Vž½»™´¿¥Í|ÅŒoDf[”µûHNWÀ,Ü ßbüʼD)©9ºc0˜«š~?ãO´›³Å›„©jܱJG½} ì¥5½] H…”‚êjÓäõvYïàÔ ÜÆ‹‘ã;Àkäß—÷½:¢°p8i‡¨;ûZ‘4„„ÿ‘ÛÈègéõPZ-¦øÔySÁ±Ÿ_ îàk´ÄJ†À•yWÝ¢v£’…|ÿ Iôˆ–($ëØ++~œ£ïý`M•­Êë{í"¸ À#&´H]½¡BcÒDuµ]΋€‘å9ƽ] <±ƒGš’—ü |í1œ"®UA–vbGD,˜V~ÏfÞTBƒaÙÈìáXuÈ+³aÀõ´a¢¹"?n¸ªœ¢§ai2áÉL¿?„¯»M4`xváˆc. 'ýäƽ©ÉöY“E„˜M=€¨ÐNù·BÀ‘Ù´‡ ÞÜ¿:Ýw=ø¤ÕÜ.O×P—º'P[‰ßmNDp„ˆrܱRsªõã‹œ“žÐ–ÛWü@ùMBmuÎÙùÿø\„ZwóÖúÉ4û¦,™2_j'„¥ý¢ÃL!“v‡çª&‹ÅZ9À ¦±ìýüv¼­Q„AÒ*+RŽ&…Õ®Àzÿ@D\k]öÌ(g£ñ_¹ 4ÿÌÓYIpëU’Ì£I¯oz˜Úu%Ï +[°=¢5|—ܨËã?o¼ÜÕl9EJޮͅZ˜§Í™×£6` ææiÓ±ãøÁÝþ—ömÊ‹8éûª\̓×.a/Y”¦uãžÍi <å4èrŽ€‡r¶p}O¤I+1p»¹h´¤{[ÅÈSzÀ†z·Asvaá#ÿ¸ÅØ}Ï™KäîÞ"šIɶ,º’áH#Íp+e¿?r¬¨œ…hÌ—0>-`˜J*Ë™2~¨kJ*©FßΦޱæåÙ*t?g.` bÂ'7>Qcc¬ÚÊV–ÎbùØÊÿD :ýs˜áеÉ7OjÂM |YÝóà5†hÂZ¸…Ën@qygøš½-D'tŠÜ¯Õ+‡\ÕÍ©Y-³öâAÁ4夥,àÅ´¤¸Î¢§ò3©Ù²>&–ôƒ?¨·wG >3µ©ô§­LæŸk\·$·È€.€)‰˜zÂ¥_H²ÿ4ÝÙt°Ki1Ì”©ÌÒ)t_Z§ÂJ.^iCAÒ:6F¹`ìånJPå9vÆæé ï˜d¾ëæ!wêsœ¾1¸‚2pÁíPf³¸ÖÐçP͹å|ã+^ ³ 8÷åø”àF†Ã4¸Ô¶¶pcÄ°A¾ D6ûcf–qÃØØpÙ ?íÒûu¤Ä(íóÕγ{£G¢2ùoðÁv‡¡ÓYœyk›`íÆ{Ó¥ÈØlü˜»öCHkJw¤{ Ö©‰¤ tÊ2.å_Šç;T/ƒ‚”‰$és)Fn$nI~37ÅÌÔŸ«f™ ñȲ‚&õš`™¢E~ÜDðMy({3-¿kÒŽS’žò61+ðÇ¢S·H;–šgTÝc´:ÉÚžNUG4ÀÀ¤ñŽ“…FE *DYsŒ˜¯ÒŒd2þêêK¸—Wü‹ßÞt•fwÞ@º¦Â íï·¡=‘™}:f«Ä"ª´¶6nŸ!xÙrMûþž¹Í†ÄhpfÈÉç´çŽœªäÑ'Ûåv±’„<#ðô4® ªqbÝI»Ï(ô? ¹ÞúXÐA߉FÏz‚ßÀqþ ðÎû)Ðå»ËH»äOôCZ¾ôÓ,’+;çÅÖÖ@$E¶å”¯89rªÒÁÒÃtåÒ]=k?u2[­H÷ÊŠÔ)ëÂî»ztè&.Øêä}Â~~S* 8ýˆ®ß^ë»Ã{HvìM´’ÊñɛЩ5d£;YÈ0+_'ߥ%tJt~ce{Œš` ×BxñàÈ ŠÖÓ $‘|ùÕÏtËu¿çx]ña'òÓ[þsبÓne±=oçÏ·ö>ÍJЪÿØ(eœÐ„Üpˆ¹x¼ S2¯¹¢7g=w²¦«Uÿ–d`B)™ p´Ç5;Ð ãä •æÍì ¶Ÿ¶Ùô&¢rg˜Cæ endstream endobj 1366 0 obj << /Type /FontDescriptor /FontName /UIRQAT+Sabon-Roman /Flags 4 /FontBBox [-165 -228 1000 944] /Ascent 776 /CapHeight 708 /Descent -220 /ItalicAngle 0 /StemV 93 /XHeight 442 /CharSet (/A/B/C/D/E/F/G/H/I/J/L/M/N/O/P/R/S/T/U/V/W/a/ampersand/b/backslash/bullet/c/colon/comma/d/e/eacute/eight/f/fi/five/fl/four/g/h/hyphen/i/j/k/l/m/n/nine/o/one/p/parenleft/parenright/period/q/quoteright/r/s/semicolon/seven/six/t/three/tilde/two/twosuperior/u/v/w/x/y/z/zero) /FontFile 1365 0 R >> endobj 1367 0 obj << /Length1 2426 /Length2 25141 /Length3 0 /Length 26171 /Filter /FlateDecode >> stream xÚ¬·PeÛ–%ŠCb‰["‡ÄÝ]2qww8¸»»»»»»;$î;Ÿ{_Wu¿×UÕÕ?NĉµæškL{oBy>+= °•¥= -='@^WÏÊ’æ—½®¹‰>€žþSFOKB"` Ôµ7±²Ôµr„mM’º¶z'3' #€ƒãSQIÒÁN×èS‡‰ƒƒÀÌÊÌÂKP0 LŒL>qÍ]@K}+ ÀBWßØÄø7ÐÕ3ì¬ ítmC+[€µ­•ƒ¾‰¥ÀÞø†‹5ÐPWhøô hi÷ `op±r˜Øô­¬]lMŒŒí?¥äú»ø;6€¼‹=ÐÂŽÀgnûKÇîMh´uÐ~úöyÿß-®?ÍýeÜhkï°2ügÀ¯Oçm­­luÿ2¥kið‰ebÿo>YÙÚQÿ%ü Í`ieÐû+ºD4 8Øýõo`bgm®ëò×ÒÂÊÀÄÐäsõ ó)Ö7·ú+®Ïèímu-í ¶¶Ÿ['{c+û¿:[ÛíìN¶&öö@K€®õ'º£®ù¿;Jû_O÷'¦ù_]{Ýÿ_RÎ0ùŒúó€Oä¯<Ð}‚šüo;=;][à?•â_ ñ·/ÿAþàO´ÿ ýOÀÿ7©ÿDù÷äÿ?§þì?ŽüŸãþTû{Àþ P÷Ó;#“ÏžúȨ̈I …®­Ù¿„úŸ#10|:­ÿW”F&–°tMð/KC«¿fòo¹ƒõ¿9mí>g@þ?æ™âï°²ül  !,”•ýguäÿVÐÿd‚þ©õÿãþÿÝ¿zø¿xöÿF%°Õñ¿I$°ÿ‘ü_Ðìÿ™FþÏ$û_#‘ÿ´aÿ/)äÿ™@þ[‰þ'ú€ýÏšã?L?ì•<þ³žƒýo±ö?%ö¿ÃÙÿ¥ˆÿu„ÌÍ¥t->g–ñŒB øÏl&F&ÆÿM_×ÂÄÜå7þVÿWeà?¦KÎÊB÷;ü2Ÿ¥ÑgÅi> þ-4±6qȘØë uÍí€ÿ+Zmÿj +;“¿^>¯ÐÓÿËÙg±ôÍ,ÿJ Ë?Ž€ŸÉÿ—? ío‡éøÉÉH*SýÓ›Éß*Bµå_+yûÏÌéÚü»àïc]Kû¿Z@ÿ?!ÿÞ3üϽ¤®½­‰3@þ“>?ÿ¶Òü<âç·rv£a`fÐ02²’- €ƒ‘ÍãŸ4õ>;ÀÒþoþ ìßö†&Ÿùú°Ë Vú”¢(/•!Þ.£23ˆöÄZirî–+×jsnH“<º"1½–ŽòÆZ†æ”Êb§¯J%¬óÚzò‹¸Á‘39^›ò"®}ßÇ\—níŒRÓs9¶¿“úÙÜ:ô?”j>'Î×­lw±’’0:ïk…]–G¶ÔÎ:ãã©$¦¸üZÝÌÊ “$@OO>ÝÉBw–(&†G•ç¨'1YŒa÷Ðùɧ4-D˜P¯ÂÅà”±Îm‡üVדºæ~žzoUx–‰ °Ÿ‡üÇÛiÅ“Ã5ªaQF³Å€“Þ6 Ô!Š]íàþoP2CQ$Û'θÏÂ_/8Ïî_­ëÞö‰„$éOVEß5úË|oÃîÙS+´! D·3íø å7UT uà 9¡ÔF{zð—Ñú {šs@Á5…M9„ÛÁ£a¿Ð£µ[Ã_¥&%ÝWõÎA$&p;:d¹FâRÅ•jÁ= +¾Méô¨Ð© „¯÷\£– Žø>Ø»"ÕÜ‚‚ä1k¶É)l>9{zý³œOº›Õ%%1•Vé]}ßÒÁDÍ̆~¦Þ8ªd&yè,èÌ2,ü©U3ú–¯5`n¬«*mÊÖÈ9ëh9óOo,î~þ zX%ƒÍi|W‹ŒJkƒEö2Ôœºd´Ïk.MvÇþñý4²L láb`£´IUï#ntñ‹a‡ú3Š’dûQƲLo>1{w‰þs/, ¿°]úÑeÓ½ÖNœ>×™!œeÙŠO4ÂLU0:ÖEIÊ5W 6Ê[ÿ 9\·•×+Z7 ¾©£Ž-’m‘%øQŽLäÀqkÇêj¯¾HǹC,ãcöÄšÿƒ<ÞU¯GÓäx¸¢È¬î¶ü{u¥EÄ,µ3™•ßèûòtŒHz\AåîVÛé)?U5,ÂeÇ´KÂÃøp{ Û3RëLlîúhlÌà—§Ð\4-eß•CQÂgìB×]Eð˜çù´Õýßñá#½‹G>&-™ñ5$2"èYfó7}Híd üÉrñc¯+¥)ÉHï­”œ¯ ‚ "'÷’°^N¢±ºHærK' é­&ÐúƒÜdî¼ö,ºÖq"•zVãÐQB¾æ&›jÔZ½ÈÙ‹›T":yušÕy_€lž]=P4ÏÁ“·yH‹u??öà‚šTç[“ª°Ïö_d’+q ×ÆmXø_.q‹“y¥çz³ùLp¶¹PüÑs‹OôÉ4º$žû^ÓÖx̨)&á0cýï”ë5\²(ôõÍŒg(ýj½%4°ç²´wôBaíá Jï«#Pmû½øÞg±&1£Ð áIsstýVF™>Æýî=ýZ5;i[Pè.Wð 4{v„p3o’0»ãHñZ^5»ÑÙqá$ŽíJKüÐ…öFé¨èí3º¾¡ÃWVl„—3º›üããž Ö!$ŠNCÔ/´×&÷£¡½pÝ¥)izYi>{QbJîÞ2AeÕ»*(›-=Z=½'–;h%á8Ä9Š” ³ ž †ãæ3ûÚëš$Cû:=gωÌã*E÷ÅƉ%Ò= ² ůÐqþy]µ8&Fe86ézÈ s<ï;b‡>@j¯>ë)„øØÓúU9}™uX}ö­(ÖµrÂÝ‘7Óg)hëõFÏ‚i’fž»‡Yæ¾Öº¶vá}ìBkK8Ñ WLD4a3³¯‘)Û?ÞĪ½DŸ°TÙ ÓÐ\AcPt5—_õ–l%XðÙ.@jÊdPòb0 ð&FÇïgG×lN”#zl¬¬°"µŒ‹#8ùÌú¸”Ç>dqÄÇÁÉ‚HÉ`jxÇfΘÇ"•.HXö'œnœ)´Û~jB¢Aë…d˅܇æfk¿ 3haë óÜMÐ1Ft3&‘Ëò<Z8jG`me¤ð<«¨÷£-me³‘-PÕ ˆÝ#,H@ŠeQZc’Grkp:X0ÔÞ3u:,¹à’å¶CµË5C_¼WhƲ´+>RŽ?$¹øÀ—œÜ-£ëÏ .›lLÍxêZn›±䈵±‰¦† ìvZ„{Œx´!£‹âÜs~wÌ›½ð¦ÑIÜ>~à¶ÿò¬À¶á<yÉš=m˜ù7‰’&¿Ï° Ó«‹Â5 S—î°W¨;èæ  kÙT,QÎÐ7’”FV,ûTqrmnèeÆ7³Óž¿Sè?]ƒ–…H¡`}Y¡ Lõ\àû‘8¥J2׈Å!Ô³Âmo tL!¡ä pu"L;^[‡.*5˜Bsö3±ô=J¨Úþ$Ç?_@0Å‘Œˆ_úÔ¢ðÉ9#~ $@_›ŸYÒ+k÷ »€èw}yôùÛ9ã*”XÎdî¡Ï 4Ús›EöØ¡qîîâõáõË}ì²ÿö¦`¼"’Ût›žÎI³s;¹0ä‘py2¯ÉÈñÇþµëcÀ£ѯJËäÅ °}Ü–›¢ Ç71ó½Wë„@’1Ù ^ƒ$º®'KuDQö€Ž&B^¨Yø.¯ÝˆÚ˜©t<^“X Gµ4³ÝÓ«@”/6$xª%>ž÷PD AQŒáµöZzFVrWÄT²=Þê Ïä0»|ôÒÅPlݯ¿QÃÌHÙ ¥ò=ÝfÕðæh»Š¡`“: “ãjW?'Ê‹ÎD—ô×Ý$ôÞÑd¼_£Ôs¬-LTWPAe)_CŠdd‹‹yŒ9›áˆ6æ N»ïÙeq¿ÂþB´–ÂàÛ€+¤JXáü1®¨b¤¯=AJ–é3²h¡ùÓS»l]0ÆëîTnñ†¦Ó˜á ¿ä8“ßîeÈqÔ©ìŽmøÛ¾![Ú¨òô²}ïÉäƒsãºaÁê£Õ¦ÌêPv)Ê°A†*  ´DjárÊÇ!6pAÐA5«g SÑ—Ÿ U´Õü.šJvüù扽À¢z!†(u2‘YSÑŽ{–û[C•œ«`cýâ±²ÊS÷ds-¾Ùˆô( …\ò-¹pˆK¥M°ýÍÉ<º€‹ÇÂñb d4yæ=þ>ê[;5„äŽ0ZRÁׇ}2g3ê?åqXæ?è÷ÜÂe]ŸñR{‚.Κï±p)L”P}çp›zq7Çy¿VÊ$tͦž® t³cŽ50ÂdÑ×ô§âák”ºæ|>TH^Ya™bÍ|Ú>íjM†$Òw'tMó¨€y…a0ÞoÜ31`´›©®y § 잃Óø&d#xj$æoà¦ÜŪ̸‰í¿¥Ÿ¾gÕgõö(@a®Í }‰ÆÐxâÆ °À¬ÎÐ k!ÙkÏHHÆöɺ0È)ªËAAlˆènæÚì¹pG¾Ã¤s­p5ÓzvãB3‚Êb7WÌ)‹.°ÝÔr›cÔšLçݳsâW Ò]» í#F†(·ŽúÞR3k)éBs{·_ܦKaŽ5’çÚf&Û,»2<“Ä¢ ;ù¼µé¹–ÊeÈŠhÏHp¦€lj¡¦\¨¥UY 0´ðEiÁ噫ŸŠåWÝa”Yµ²BK¬ÍÀÕãÕ~„¬DìzR4Aö,í¢¹‘ïÙâyI‚.‘V3;ŒoÎ{m6\fÉÌƘÓê¢2·aeÊ–îW2bR0H²–KU¦ìÖcþ/jæe:Aý]‡ó»ÔºÍ0õ–l$®ÄRU,œn–³™Mût‘^Yu”εq,SBå}“=~ÿ·%ñ"áoÅ¡w‹~¤-+?˜Â7çDs5wý®S\L>\Ø‚q݃šï¥°¡^í:x´?ž,tK|†àxªŠ€¥nÓ ÐAö°§QF£­æ_ܪï þ|ó?8ÔÓÀ±5ä_÷J™¼ÂGyv#µHå jÕz7?Y÷5ïV¾ñòIt‹Ç\éÚ3Ź¸:=%ZÐñÖHåýùæhï]¹¿Ù̽Ýiz³× Û$’‡a,ÏÖã%h%?Å”úè•Àħ|dÜ“¸Ó·-Ÿ ^Tâð$Úýš´•ãÙw½y§L³¼0z2'µÏ¥#¾"Kü(9±«€#„-¡5ÐLJQ³ªUcâ-'” C¿JõºŒmÒ”fwÉUS«:“€nHdÖ}áÇ ™ª_ &Zsbf0nNþÎÛ07ÛȆ›û‘Ë_¬’r õ"ôÅ‹Ý?ìêMÖ-tÅ¡©è)þ<^= "übç¥ÃE›;E-,_RëÒÃè;Æ]#“Y…@¸Á‘éÔÃ@yÄÛ€úèø‘1hEþ=¦ÏOuøC ¾§sÛ¦Eý× (¯éÇÙIPÑÔFƒÜ¥¾®áÙ¾"GÐ8±¬êí§%#µJîž“áÁzŒ®Ïe)_±‡<¨x –°C“åa­ýG›Íè==ˆT…py‚²ŸÙÄm÷[éƒA©Q<ÓY«¾¤¼ß9Uî΢,®î¨h=§¼eçÁÍ6KXõÓáMxžÉi ,ÃßÑUt<)mÀ\AïdCLàO=óüµI0/\ tŒ´¾.ÏÍcyÍ8ºƒ¼0‘Nßqðl½æ W`‹*¤ïËaŽ±ü½Û†U<”ÑXáƒ:КÎ(4¸·+ûVž´ÿ•vñÊbœ¶¶»ÌXõ‹÷*Fn@èkPéþ|ÀÛK9ØAéÍ‚}˜Q70¶Âï¬ægüa²'—Ѹ{PFÅGºCðÛªjà €*> Tà^²½4ûNË™Bqòûd¤D=WlõO9dùâã!/ôJ뛌ÅÉá^ó5üßM«»·ºì´Ððá\~¡§ÒÛl¬Š²KœÐ Çö1aÝùÀV¤Öøâyõ¨wS– ا£=ý`>™e‰™J2£‘̾Éån[¹ã@)r¨dßœQ|Èj½ø™0Köjÿ;‹eç·–r3€S6rÇ°2«ÅÝ)1q¨ú´Á7< Ü~–ž\‘îJ4:q»Wž¿6žúX¾3/}ÜäÆžF)¦7æ<ª3sàtoω·âÕÂ<-+>°·0€lÑ­B¨‡²…‚ÜTæä‰O-Å£]ñlvF]½¸YbÉ“)§ ñÞ¢Æ$m‡ÛËœ;ÄÏVj½­k'²›$l¡qÐn|;pÝ•Ž Ÿ…_=˜»þŽš2 f1dÒ‡ëÛo·gù›ÂçâÝ™¢ Ô.m§½ºê*¹UzðI~[%ÈyZ@wÊ› ¸Ø¶y ä§æ ôÉùVJø15»õdÞé)X˜•µ›‡ÄÂy>ý=…sÓÍký€h—Y‡‹q q.ÄûFaE´›ÖVrm!6Ò;rÐÒÄŠPO”øVFˆncíñ|Š$´ÊHaÓ¡z馱éÂlÝ’XýÈÆ™¡à§þµŠ‹>³›"»~Ȳ†é´l’ôŸòð3$:l•âoÒ j8‘ :ÞF, ½Wçi©E¥Ù}Þ åŽÂò¦|F±¿.»x¶Kà%Û$‚+ÀîÇ”ù&Œ“7\Ä Øþ< °û­î˜>«±m·“,DA^þ1I«ÿÕõTÈ)¦,~ÄÕùÞÖ¢™éáÊúƒê¨õ,Ægéº2ŽxÛ4X©Ä?hÍ!@°`F×Ó÷XS¯@ßF%}N²!ÿBHÖ³¬ë4bf¨å°BŽ$‡Q܈·³ö3Ò\?5fßÉÏ7_øv Ñoåâ—{åi±uëÎo|¤|¬Ø²"W¡°"â´o!p¾Å¾Ö@µa .s¸aÜA—`J@õI€ôgÝ —¦cÃï .»-GûÛì÷$Ĺ’ÎÍgçßV.ƒw†No]F6”ùp*ѯxŠ´øe¬wû4¡ì››V~«ÄÈ'E-tz•R½Ù¨åj?Ñ!º*É&+ÕL„ ù]¯O™ïã8WlÒ¡« Z©·›±šq>˜«Ì:Ä¿ë &ÒIÄÆ$w8ŠDðhÞbsA$œÃÈfi$x¹êo&J€ˆžj1Ìᄳ*ôͳ·ÀT¦ÉkŒAîŸëFª:„´ÃCoPƒl~æs¹jÍz«‚LT¢,sPG‹_*þVçÓ÷}Àc£‘¯øífQá÷ÃØbjËŠ¬âHÝû…¨õ›¾ò0.Íafß.ÌeΙ׳ â%¿u÷à&¿D³Ü€-]߯v<üÉeÉÜÍ4öÇ7û82æh¸Ó±>DzFOÖh™¬6ÓjëåN^ô ¤“ ãÀsÎET ѬôY8ÊD¤}¾ïÜýI¬–‘ÛWò¡„iˆ‚ðÇ¥±ƒ/ÄÛŠî4iSF97AR=€ƒl û7N‚«þ4š*¸²|ëå¼/¥)©¸Š6PY5Ü8½ l‘ëUSØüý_pàyÓ,Û»7M¯¸6 #fo¦æõZ[ShB1ûãŠ?=~*v§ñKmbÚaדx®`pä$´¨&ª ÇyjM¡Ñõ{í% ÏŒütX2a@3gŽ¥ëÄÖ`ayeµo.lõa¿²,ðs1"Q¾ÏœãO g‘ðòf¨§°OÌ™:[W­:À†.O7Ps´²wˆ\ÿv;5J zø&ھЉ’­›˜°Á~«RiEÿTkIE¨ÃF£î.é§ßßäÔ»ð3Üt›¤@,ô {ÖHb&ŸýÃêæC*ðHãë\†pMº—)ÁWʇ±V+ÝZarØ„Ó—¤ô† Ù¥…Ja+NQ,f›äè|Û Æ=]ôž&¿{¦¯ÎÊ$lBÙ•nɱuN¤%"p<°`¤Ëé³ÐìƒBSÒ—¿wüž=Áí˹Ë'ÀG^;ܯ>¡! þ=GRTLôOÌ.Wm6™ð)|»ÅoeÙ1BRí´&Ö$©pob_Nz5í™éôyû²0^ç`cÁ—(›’sZ‡†Ú/ÁëeKÑ„J¨pÆæXî‚S±S­j?A•ùýÄ“PJ!å7R.iÁ‹êÓ¬˜ °¥„rßØx¢*•nSxkÃØz½sþÜ÷lK«Ÿ< 6ÐE;y(6å,±ž†°ìQ“Z*Χ6@Ì .é€ø“Cwh³6ŒÉø¢ªHÉÙ$ýa·#s’vZ?qÌ/ãûû·âf´«eÅ7³å1ù°j4gKâüÊÇYo›c„v®žoœoÕ6¼IͨÖàœ~×$«vb$Içè +CÐNT6QWëÒ1ìXA$¹Ì_<Ÿú[©S ?ä¿(MÚ*äùú‘5áKbÓ†YJpám©ßoø©û®|‡5™Û¯,~ߺ×Õ%ß¿J!ÐÅ=#Å#Ú~E¡^ar³†%v,%›@p%€¤ñᦤBgXŠÕ.òI'Ÿq¸tîº~iú¾yJE»¦¼—‚Øü›™¬eCÅã@T!©Ö0×ÿµe¥{½ÐZ’MTÿ¦hu¾e©.k ‚ÀU 1ZÏÀ˜Bhʉ ñg^Ð'o°q÷YñãìÉ~ L!Œ•E2ûm'ê!å€ö«€iµQJE`ÚÐ>yØ–§ òè#l¢£);wþ&º˜©¼®hgðrN#‹\˜>¿0¢¶LÀ©‚4 `LÈ-ÓŸxµÌ墫/fbUœJ‘Nò:ˆÒü†LSt§S¨¶Ç4Oí,d{'é„3XÐÇ0XšCeŽ¶ ²6dñ³8µš³ŸûfÓч,tÚá)oûýòOi=Ì—ˆw«0ÓÖLƒ•ŸküšÒNÜ"¯\ A.ž;…×(]ÇH»`mf‚™âO¿)«Ü¶Sx&ýíæžÀ~wÿLé¬28Œòý6kîh÷½Ð}§ãý¤ö`{™\ßÉx!î¡Zë^]åu“Èp6÷£éÅÃ4Òo@©÷ŽSìüâ;Ùof{†/¶X±—­œI¬bäuÖèªYt?˶#×,›°`Ìt¾(¾Gížéª,ûw­Pßþúδ^ï¦÷IÈ8 X?–|vÛMŽZín'6ȇ?Ç ûK¸®ñkÙ/žÝÞ)Ë>>_„?nNŽk?.Ô°pÙ ª0¿—›rºàÖG¹Ù¨ÿ}Y³&XÇ+ÔÈÐáeuÕ•`à ÊL‘ËNïÉ)Ü5W8aˆsœÁÛêˆã¦ïŸ5‹}¥Ìk{%˜ 7))'âùµ>ù°÷Ò6å´ûÆ Äƒ;OŸœªm@3 ‚y*o£õö×rëf±!–4lŽ¸¦ÕÁíªìq8~yÛTÄmz§„ë¨=6¬0†v'ð'ð½@l«i*¾qšÖ)äôÑPºÐS“æ§Ëg¨œŠ¼:yµúÄŸ$½«¥è§EI²ÌÃalÈóÎHb5ÊvB<;—T4ˆÎRà eé'C{—Ó£¥×T v?¤X{ÚÜß¾>,$:qî_¡òtU0Uñ¥L›{-j©øð³èÁýSÝõ²€µÜ½K@~ê†ÛXÑô R©?ëjÊ‚½5gË%ßC‚tþ] ä7CˆGªDÙ‹Í‘ÂZ”—êR¨dèg¯Ñ˜‡ÍùöxˆÂD€sñ OEöö½§ö¤©™©DÆ[Gâº..ÉùvôxècPS“×؆:ŒPTöèhnÀÔÆkBíüŸA…õnLëâ@¬MgzB¢€ýÓ¼ÂCÍÙ؃ÉÔ&^q¨å2Ï¥û…»‘(ŽFJá/±Ó”C®‘ºQÔ?ʇ}2Þ$‚ónmpìjÐUƒ°ú2”ø1£"èÀl9éÃEco#{Æ(¼qÂÕ_­S‡iu꥟síÛ #µH×ñ¹–ä²1Q£$Ú± ÄÈ\EÄÍå²û¿{ØUý#½L”™çE-ŽðÂ0ׂÉdøV´1V«°k&[¹û‹éúçÏì®WÙìN4mxvãkè®1Ak«‹Ç³ëå~¦‡;jY˜Wäß;ïÓrgÕM]yb°¬Ã–Ùd™ùÿiWq¦å½b‡æÓ¿Œ¦ÀC&sžÕEÀÄG8²@¾ÄO‘._‡&á÷Ø.æHòÞß»·¿/7zYeUðerJ‹ÏUôN¸Å—(’©6…™Ny%gð`O~šDÛoçÇ.·Uð7þ˜"4|b2'^&ȤDNlÐ<‰ºqó쫳§¯ Î{”e\<ó¹·Ú.Ó·"ªµH³«©¸\N€ÅÖ6xcü2p›øÃâ‹ d7}"¨ $·©FsMÒÁ·aLBþÎMe„—‹?”ûî7qËû”¾ìbòŽ :M ¬OHöPݸc…B¸IbPD©fxœqi)'cß@,÷ˆ&k±™m2]:Ò æˆfËÁçðx0Ü]U;'RVÑ„´”9? ï{E%ˆkug¨°÷J«UÝ”u’¯2GmJãË[p`@,ò• ×…BU7`5:{¾^(Ç{%`Ï]¸>¼~<øélk›i ºîþ‘Ä„‚ˆƒÿæi ÍOŠœÿ„XÜq6ß©¢i®œóðè¸Nü]«*ÓŽ(Œ×è+øU 湤· F=ÞÑwðÒ½X2ø"§€}¯ßhà7ÓÂEŠ¿X¬V˜{È üR2óBÎù0ï›=tf¤V3Œ‡¨s‡¦‚¡bš#‹¢E‹Å=]~„‰3Î&ç Ó&¯Î8à\'Ø£Øm.D—HOu=6Ñã3þÑ/{£h©¢% ­›Ë);¤…Á`7ÍTO´S™¾ ÄþUŽ¿~?ùiS 5ÿâÜTrð¸7œÝ>(7 9IÑ1ëv¹!‚ÐŒ“!TÓh Ú6ßøÇIfO³Ä OvP~¥á%陧¦E'̉m{[Àfá'ß8ÛP{ô÷ô`NìlL­÷·FPñÜÎÚe;åú‚dÅ«ÄéPG|=ô‚1™M“Þˆ8CH—¾7RV:óÜx;s4aG/–1•ÑîĬ—†DKÇ^7,*—{Cwdï–qdkð³ø¤Ü“²°L/‚EOöG¸O¶[êi"©?Qû¥„¸—Oˆˆ“¡:U2Ϙ1™Ê¬ž‰×ÞµC—*œ•ßp»ÊüO´ÖûkÔaí¼ 7ÍlE‘€U.ø!²Fè#ꬿ’ÒžÓR9û¼ÒºS¾­¡¶ãv˜*2ëƈe “$ÃLðÊãe½ýGñHï…w ¹ÉŸëí÷:öѧ}eBNÔ§@ŽyÔý­fR´ço9rŽDîãµòÖ÷»„‘ãZSH%iAiC[wÊ¿†ÞI{`ÂW„ÆQYSm™M~zÏWþþŠŸ½U°‰¥fSYü. Â)÷V»í´ÁZ_°6™¯]ýÎ6R›äH×Ð'çÄÁˆ^Œ…k%ò,x9æIþžZÙÓ塤äkæx£I¸ó%âœâ¤ø!:cLô‡A[ºZIËfhæ¸^ÁZ2â|Ù:¶¥ =hLLNš3Û銓íwæ4 W5‡ÈuFhqnŸU;¬âxuµjFŒmXnl?|0K±}„Rg³d\C³ÿKm†á`vLLôë=^õŽøð–â˜@5#@ÇÛN_˜w£Œ>š E<¯Ø·½ØDJµeˆ`8ZU<®_¾:Hý) 6ËZbn¡cè®·p€ì!Ih\Ñ&ÈMNà ªÍÕ†8EˆÓ™âÂFº)¼P3[‘”F@+Ï{„ߺáãÚ”Nا?‹Yk_²‡Ós¸Ü¾¹—ÀȽ”0Ò4Ü6t˜ Y4}ã°ý;7¿Ï4ÄŒø—~U‹˜;HhùŽ‰Äòa’ð\¨25${Έ¤©ºi:™îÁ竧ÿgyDœ 4¶hӯσþ–Ò8ûéèQ{©/+J] ªd|@3ÓѶŒçV»Ð+j}Aç£?0%CE˜Sy{¼ùX\L²õž¦ÐÓ¹š®ç¡Gè(ÍJCþ9f‰M-“«NâòÇtá1êË™$Ë%‹ü¨4¦XŠÆÀŽR‹ êáD::Œá¨V'ÏWâó¿“ÅQRü“·pN°{À³JTƒ.~ÀÛ!ÜÖzNšÜ»Ùª§Ý¡u .#°’±jO=°Zûz´ƒg??·¿X8Ÿgt·xE$mÕ#ÖhRÿLnq÷‡–”’I(ò*,/-bí!oÍÏp‹a²Ÿ«p¯ÏŽ²©käíã¢g}ù®N:ïö»B¤Ò¡ÍNÛϾ¶*²½€•YÊ¿q,j]jÕ/x¶O“̼·I…C ºŒ:ÈŽ[¾ÆCÒí§Âºô~DKòà5c_3ËÇY• øö°6 ÄîʸvH²žO$¸Ç±¿dz¢×üàÆëùÚa²=É Ë£ùë°Ÿ7Úż\Ý9œ]Z™"5‹8~£½¡ká)&U¤ÌdÊ·¢´Æóó 3­ðÑNiËúøX8X‘KOG!èí¯o]ëùÐIÌÔ]œg¼»’Ôe)ê~šw÷ãQ-TLtÖ7ÕçñX:P5 ºN ½h»Ój¯îKìa¬Ì¸\š0Üðv³î…¥[ p8ÎEú€ ±z‚¨,å 6Èô1ÚtþC*ô¥^ˆSÌãúº…HHÀÑ+ÜæL%™AÂiOªô$K„ׄ‡@ÐÀ6Ñx˜õ¾îL¶q_;§Eå Ø7ÏŠ¬!î°e|2ï5ƒLÅmŸÀàý‡» EÆß ¦)œ ÞpaìDsSÂ:ß›,1©‘û=R›-fD »žábÈFYT”Úq1/àÑUSÆXGÆu}—ÅBÕ«r-ShŸW¢e‚±ü¤a/ñÄʼ÷ä§ëd8¿Ž´üz Ž7&ÿÙ`jÕ±YÜÆ Q E9®q4m… =g›fF›ã48ødümÞiZ£àëOxÏD®íh…òDJìßAF”’HøßV÷ÉÛï=¢Z3êצ¨Î+ø¼üI£É²Öú æ~ÖFÕÅZÓ³ZÅV,ûͺÞLÕËizœo#åuTˆÄ è¯iUâ,BÍéü|ŽX‘H»zîÚތ٠w9¢¹²1AæüøI&í<\oRjp÷3' G6u^eÁ†²`þÛc2¾=‚Ž½Ó·o´ðÙª‘£¸ítÖ5ù8É9Þ0â"ôȬñ¡Ûª±>@Ø@¤hTÂÖwª*)躼oÎf5¥ÊÖ,ÌÑH÷#ƒRÑçPºGä¿’ÃâÑé•EÕ3„tÁ*Ç°³¹\y«7¼W•¶#ÙšsöŽqˆf)†^¡R×Tìì(²õýløµ¡çî2”ÞõUÛ#ó{ÞAÓ~Äl¤}¿¬Ž«‘Xñ%ì˜9ç6E•Dò?ÎÔ:ÿèìifeÂ3ê¸dâÀ†™Y{V8£½¥LŽ°Íÿ ha&î²ÌSÜ-Ñy¶ÀÛ3g‘ÅiŒg¨ÄÚá éÿjbT÷Ë4M]$0F$ævFUÙœÊK—}pÚ}þj±ÿ‘¶BSbè‹8fy¥w:7Ï»&þ)q€«iûlw:Î:‹ÈôusÅ䌜 D"üþ°dS1ï¯ëì*ó7¹ósÆq®CGÉe%/u¤™YGö­ý]ÇgQu™NïZ¬|¡ƒŠšQ¥ â‹Tâj†IËd>aÊJð0Q_Ik7pôËha–$«š3@»ÉAˆ]Œ•ž”Ѧ5Hq8O°Ür^I#÷tÀi?úÃ+øf©Ë~8êÇÙçßÈ2Âk'uYjz‹}ŠõD¶!äj2ôƒ0–7.C"OþCÂdNòNxÚ°# HÜwSÕÏüéy›ßU#$ÝåȪ¼à}Ö±>Ü­2’â Òâô_“¹ÅVŽÓÍVù%4 èÂü“5}»TÀ¶¨CÎÍ”–w®ó»åVü©P§#ßð¸µî» ³ãL—}ƒò² ’t=¨Xtë ß A"9 s1ß°ëÕ«ÞÖðãA‹´’{èc^<Ú8d¸£ ï?³>X+DQY–r±„Ÿì=Å •VHG)ÈÚeâÂqfqñ¢ÿzuÝdž©<ôLj¿1Nã‡)¡œ{ë˜yVñ1R­9‡nØ ‰ÄgËFbnJö›åq½þ«êÁH˜[7Œe„¦7¦ÁðÀ·œéø‡!«¤&VK7ìä—ðE!²2zu#hY,ÔÃìÌ.çÄßjFÔ~\åUUšUßú/qÿÌ溥_Õå¸ÂKZ´¨gá)Y{¡·Ñ2Øà†¥è°½yÁx9Û¿Èã*å„Ü/>}oV5Œ[_’ÁœÙ‚Þy5ô·§JQðŒ gòËå´ÒŠbÿ(/*€8ÁÇŽ%$Ž*Ø¥ž»«:­öƒm¬aà6+Qºå¬¬ŠiÕ¢ó°K«žÚɪ›Âg}^¤¢° Á)»¥ÿê+å"?¼´ÙžŠž×óËlQ?Dñ *Š9˜ä c 9‘ù(þó·±ÜésÑß$’©|$p2M‹‰ôZT#€fSÃÅÉ%ð$Ïs¡ØN‹)‡¸~TP›_ñ:X÷­ßŘÈÚÆ#”(Ô5理NàµÃì`å¿åg€ôðÉ•æ5|pCXe;o×b=åkh¼sæ'ÌHІý×¥‹Ûð0½Ó}ær­š¶º3ç¾ä9°½£~¢Ã¶Kĸ¨4¯ÎýÄ°à:ÍøñBª2Yˆî“VüPº•s„aàV}a]¸ÙÝÀK?¡ªö•ÚírߢäLRfšÓ³U¡uŠîA£&¸Ÿ×úRò“vé8vÚy{é®|¶öÆýgµÂÝsÒK9©K×V…dÜÒçtwþS«ýó½Äw5ߟ$ dx»ØóÎo\Ò¢¨Lä :øçxo—‚`a©Æ_Ðú$öô\bc🢉 ƒÃ€Îžïrc¬BmOr>Ý>*ÉWı6 *a™ §i”gØz¤ë¬`è Œ£ö.} ÌíÉCš,Ï~úk¦¾(Œ”Ю÷(“Um¥ð‰TŒ'xøñŒ=dÜÒýZZ˜0©ÕËË¡Ï ÷éh^$ÔVCaäPÒ!¿kõÑYiç[ÎÈ/²a’,H?ç ™dŨL¶gò8t½›˜JÀœ¤Ä-«#E84k'Âx²ÌÐÔÁªëÜÑÓ³w•'Ru«Ó[É]9‘!Ò‘š(ZqE4{žþ!“_«Ù·D]ÊzеÉl/ĤlÌê‹ôæ#úÃ:Ý’,½tZiV–!Z¦Zb‹4ͼ2JVz¬O`©¤àŒÒ#ñ±ÊPõñø[ê67ÂÊ=è ŹSTZÿ\É…ôÙîÕK˜fÈGÁx>m¥JäÔ&LØ[è"ƒ"ïöØ?m£ÖËɶÇl>ý ß-þgºn…Ë®sºÁo}£‚­¢Ôk_3œœLëœeQ%Íê¼P¥|>j.¸ªóøAhظàÛ„Vß…LÚF-ë¿Að3i B¸•Ÿ8ðs@Ó(â))¦ã{ÌÖSR¾ÆÓeeB‚èR©é·íQÔË'Ôd<gê¢Aæ†F¢ÄÆy€; é€Š!\R·&î áF%àB¥Bå{† <Šë£k ÛÛœ ™}_ð„sžCÝâàó9x¢¡v×î¾ Þ·0B¬j›º'89Æ¿O¤Óñ¨ƒ×}¤v °Êàõž2 -_v‡Xª•ç¸ ~u+t$ªN ˆ¹ÌþµV¥¨¡~µ~àã ö í…Âxñ7¯™CÒ<—S¬àt­Ää—™æ¸ÓJÁÌGòifeü3$z¢p±È rƜᘂÍ,Ò0«±©ç^¤ÕªwÇŽ?ªöKéÝoP¤›—JçÞ úZGÍ€]‚N.1à xíq;ða蟀«É Z«›à˜íZ%cƒ20§Ê1̯l&SrŒ…ñN/¡Ñõ8\©Æ\A¦§ÑuK¶Ýb‡"9õå\‚ùi…rªÆç´²ÑzƒüDѬ’YTQX€}^—Næp':©dV²q-ÞP}¤1§ È½v’CÚ³z¹†!§eñHô\H¡•…¿Ž5zýŒ#'‰_k ãå£h”{î[}Ës2~Y+!ò^¹YÙÖ¡ÆàÃ~’Õk6›Ù(ò §mê2PÀ_I»„– `Yè¸SGäÔèÒ¾ÔŠäbÂisÇœ77š}¨£“O΢å&p4¬ÿž¶v£@<á­lôGŠ¯çÌ|,·¨‚D C“LBç,xs’Œª'b(˜S6rÝ¸ê­ ï¾b'VÀ(Îç­ uõé%¦tïêïåp¢‚Ýšó&OWØã«ø:ýê«ù3îsg²,XÁ ð´¸ .ýìçJsŠÓ¡S¦&¸ÉÉ>¼j¦02wìó Ôa‰#tª&.aMSd4KŽ­äùóš¬jkµ«¨-s„ÕÄO+Ýv” ¼sþµ™HîN««ï»Ï äÆÁ„ìÔè»Hwþªþ‡¸~Ý&<À%}Çw gì÷uŒH ³vÅõ¨œFw›WÖ¥Õ¸]"ÖlQ>gp{¶¡—ݤ`Õ»€I™cq2=XŠôÊŠ­TiÂ8]ËëCð½PKxÊrZâ¬ü™FÍ…èÂuv0žÀI:¼æÒÃþÙò*ræóƒ±C/P>ù ©7DrX9ŠÛxKyÝ G>%‚PdbÛܺšüÍÊHÁéjº6ÞWÄg<˜7Ì!¦£DP:yPkaÇ‹ÉÕ­¨I’ßRøm×zÞ¸] ÓŸ§5ŠUhK­I‹¿Èر¥ÎS¸ë&žyçAŽ•âñkÌ…¸,·™Ìϯ‘¿H€IžrJõìa›]²á?\R)h«ºÜêdë4Ú&¹ú!/m°_££±é0i [é…ÙˆR¿,{bXúf{ôÓÓ˜rö@>ÜÒÜ­gF#ð¨o3²×ŽHú#~Ç—£é3íúvÈüÕÞ™-µ Gé1—…ÒÉ>˜²é¨UƒØ\ƒån²_±»÷bÈ/rËßp~D"[ˆåGüjœ¿¿È·ó>¡].X8š°A¥†Çã95`⤈)Œá±Kr•ZX…fFû0O"~ËOÒ^j9΋–x—®kß— (`\OÂaÕ»oEï¢ ¢*i©³”’ê…®\JðnûIwËÁê²’M–\Ç=[Òi±§¾Ê è…©ÓQ4¨ —ø“c‰÷—‘=ù<Ì#r8_{Í–0)ufùo‘èE;À|kŸ¢ŽãÚYJ)²„d“C­ÒIùNV¸Ǒ_•Nl˜#ÆÏJªÈÓÒüÔãrgׯ”å”#Èþt–Ö¦§ã<»‚—e~ß·–ŽßZ} R¿Gc8`p÷*×å8‚U°,ð7ögxCóFª%…J¬•=ÞÛV·ñ9µ¾ñ?PûïïkóuPlM|é–5Y½GÕ™“«W]Ròýüœ š:›3}5Ê öU6­ttb;„”‘vªºÉ=CøØéóM–hvÙæYQÑZt%óʲíögä¯å€"Ò7x>"O|·«iøéÅ•ží¦áz y`›æS^oØû\T­T¸‚º, Ç'gÓÆ©œ{¼œáIñÛŽ” —lFeOJÒ©«Àym}è‚~™`•²?@APXtrzM) Ž¬…(¿Ô`OÝ¢ö²K ^I±å ¶E2î„·á,)â[¥ÎÔâo>„ª^ÀÝAíPóy\[hö–ó©W…3kQâÙF²]3¢MŽzÞ#¡l5î= G?²=»Ä„U‘eF6àÉÊôP'ÐÑàiLa™‡2:»wÿ±\·ÿÓ–‹ŒA¯´`;ZI»¥ÈŸ:}ÔƒÄÜäwp¢%çÇ»½`,D°elJ E’§xâ@"­{”~´Wòë˜÷»Û]°lRkw¢Êù-ÝkÏçJMVD$)d%½§•ç~Ú ç+î³hGÜ~rSÅÕ±ÈÉYçàA‡®P;ß9ô¦H…ô%P9¿L…0B"AtYèñ ™û©k’J¤P>y…ÐwYCè]¹”ü˜Ù’}›ô-óê$ú íÔNNÕ<¨¾$¹ÔG: 3T^î9g¹ýÈѬÜùa»s ÃWú:À>‰ÙµÙ$–uÑ•‚šL Š{ðîU¥½že;“w¸Õµ²“1obG¸¤{«s*s²RSkW劭f¾…9ÍBš¯{ØÙ»FŽ@ùbÝK¡_âUCjÑŒÓ\øzˑ˱Œ)§çQDŠ|5/šˆG¶Ô)ðJsm‰Œ` ³KžYÒ¿‚]:Ë€yãäÓQß´¢ê1^Á÷U./Þ=òjäkµ’ÛM¤Àë«vÿxß+­L, ¬\Þ<Œú$3ì,÷ð‡ãþF‡)Ù®€úÚŒ¤þDØ«'’áI42ªü&XG-N³ÎÖ\X‡ÆÿÑ$.Û@wµçÁ¼G®Uñò7ð6¬®Pî=ÁlE ¢>~ûa%$ûéÞ¤dµ~C2óÜ]^GåeSÊŠŒw^£dt°xî>hVMÙvqZ³Ø{¨.#’‘ó¡`âMy…°Ð]Ô`âéç4g8÷úŽãJKMåvào” °:à¼ë‘¿$@–æ¼’þ-[ë4cIC<ŸÌ¦7‚ÛÊ*¥!q˜Â’wv,˜µRªÐ ²¯.aÔˆcþâFPk˜L%Ùû\("¹¢Ì³sbMŒÕ’,žZ„(š&Ç8Ñèx-g ýÏëã`NßµÂcšýsè—®Lè ÐÒÚýÆ4“عï3qG¶{E1Öè¨N=mÌ™BÍu©Óªò ¥œ¥Z‡{#ð±8(W”­7¿+1KT|µÜUùéÅAV‚2Î÷7?çs¤ýx´Ør|Š›ÔæKj3Ü’Æy´öýãòUÑÅ­¢iÍèõ|ò•Ñä'›ej:¢‚c˜ìn„ m‹r»ƒ±ú«\ìs\é’8Ãþå|Ô‚¢Kã1\‚Ј&ch[N`¦Ænµ'¯ŠÚ-ØØ qd“Z[†€ƒ¸g.»u$ÂÙÞ”ƒ?«‰ õÀaa65Ì Ñ7ñ‚øYõøùS¬åß{[-ŽŒ±x”‰$žE§›ëU¬GOjY­£H»²ýçkX¯ ÖC à!_œš–œ 2†¡~¡­ï7|xHxr|'üÞμG¶Xk®éo®Ð>GËÊc¿¨} -y-aÒ•`êT«ˆ‰Ý0Ð>‚ÚþY„+šÃ=²Ï‡rÖ“4>Únõ¤.%Ðò1Œ–ZÒðåo”‚¸“" 5£3EÇ0ÿìÚMR¶ÄqÈó0°ï4k¹ª×¬ØÉ€ÿ­Ä;êok˶1jÙõâ»@p$ í[Oøàv]Xü¼¥w È¢É×NŸg…s!§ Ül^dMœè)G]BWœ~Áje &?œ°"¾2³¿Ù¯™qçD}l›TŸXnB†¦Z_sÔïßÍïEö¯Î®ˆHwвD{]ëô_ý.¯ºÞ žÃU Ñ0BÛþòÞ@/Ы€ *›Y¬“ô…õ _×¾ôÏ°{ú£¯½ëâ_÷vñîŸ7Ëšð¡óÚ%ÿ‡ /úm8³N ûîZ¬oÒ³? ÷ž²ên».óHX)Öß¼to-ÑÉ­Ö½Yì©#=»Ùi$¢ÎÙe/Êg#T«ÏܤL{´Úƒô§d¥E÷*³fÃÑÔ>Œ{ÞA,´ù@ü,dô™æPGÕ£|ÃOáe¥lCq¼R†aع+ÛA³… Ê) 0J«µÎvèðúØ¿Ñ[&¦ükË^°ÖÕ~ë¡™ÌXM |Ó›§ÏTƒ×õhŽy+1¾Ñ{/šð%Ž •)­xÁÙ‰´Âÿ­?ôd¹ÿ[´R”î™ àÌbÖˆ÷É[ÈFIiñ#‰ñ¬Pæº^<õÖ’[éír”¡`Ô a5ÑvwÇè2»/}Àƒmé:Þü›Až0ÓZ‡À?ʵkLÓ.çT”Ÿb-s…åèg„YÆT×v*DS½ZdªKOÍcäZ Õ|n3‘}]‡ÂõþÝ+µ)‰ÙNi4Àïº} œ[8•¾·™_¬+®ôyÄzò@lë™ïëÁ«àjØŠÌâ2FYìpüˆÎU”¹™ÄlZ}½¡ŠîÎâ/Ã[^}BIQÜÈì^](­g¼(2UdûÜ&ͨóŸîÚIbZ-r£ñÐà¿ >Ž/tQp†©ï7§õ¼~Uú°>s®¦@2N¼€Ÿú>X ž˜{Äν\|+ü"ÿpƒî)7yÎ\`B™ âl£ÔøHpèH·’Y¹¡XyüøÀæ>Ô¶Ù×!®E$ }~ü ‡Š:oxbƒ§ª± a1Òª4ZýWòœÄFcËÑéÞ3õ+,žMœO¶4ž\/Û§ËJÓ˜(Õâ1[Ç1 ÝÉÏÚ-;&'æäg0Í×ë¾syÐŒëûÆ€¤Ç9.!z¨Ý!J07:ž~|éKðð×áù5¶f|Œ QQÕo¯àÑítÉÿ®n]Ë=ÌÜë\<­…‘#Ùâ'È@#Glúò~s…?•¢&·rαˆ<Žpºò p(]Æ. ;emGj‹”ïö)Üø6áJãU•2íÚ9åGý·ÎÜïÉåFÀÁ/LI6YG¦äD}Þìhý*Ï)a"$„þÕõú[Y'—¶¬¬­ÈØLT«óú`R¡¾u†çTHæѳQ båi9YŠ#–µ8A~‡¾V*×0ˆÙs{"ôÇ!q4 ·Ø™*ÎÞb/ÅÔ>Á ŸÁ_[¯(´ê±F„Ó‘Â6W»~ú^&{yg^ܤR­""9møž^ E4s£Ì {¿$2±­+žOÛ¿@]×ìD²™ˆZS´Ø}of$ó¾4vƒ³Æ¨¿Ó¹Ãߩø™D-Àô%olŒ®+Yãn9eÌ Ü1‰øTÚ+òÉKA\ñ¾it(/mEø°s–#_Y{àÊÎ9ù9ËñÈ1¢u½PðÆÊ–Ö£µó– ðà­¯¢Ae1u2'ëP}§÷í[*[}®)"h!¼cÁ”¾ L{ÛbFè¢Uüv¹¥[±JÄ,{~ÏÀîÒªd’¥9¼‰Ì:ß•;D†RkŒwž½ŸäËNÛÞæ¦x[VU  –†bÛ´p;§Ä"l…Å%¡8ö¿ŸôeÝø²ÛåØœ\Ë Â^a Ÿ*|W_˜8áÞ$eÞª¾á1t­HÄŸ•@È çÎÙ[… Ñ=++ñ=¤,h¢P@Ûv¦ÚÄ3è_$^‘­˜Ç©@ûì‰á¥L­hüÛœ ßðˆ+#­2–ÐéœHª¶Ìk,¼*™§Z‹º‹föçæšq:*›ÊTÒ¢îm2\mšx5ÊaÙNé0–Èœ ^³@%ÚîœÑ€öÖ<¸€GÝ’\˜ù!Fd_UÙ¿ø[_çK#»€’\ ¹É'’ð0Wõ'\ð;Ùöøðµ ÒsºçAÙ“}ݵL›Çs׊i¯ºÝ)›QM¿Í*!¿æšƒG Šcìªö0åõæoMn é‹SK¡ 7â¿•!Í­=b„<¯>„ŒÊÂ?°)²,“Ç€*V?hrü|òí+†´îòñÃM¤ mÐ]×¼TÐØÌ)lÌòýÏF‘±-¢ì·7 Û2ê³bðòì»>3VãXŒl¦zB UýÙrw¡ø‚aÈE,’ŽåÒ F‘ÇÜÓ`qV} Óðh`§MóO¶ ©\K¾[ ÄÖÙ'B¡ q¤8aYL4¯m&цoBþ(¿×x;·£)FÝ Å¿ð[ô¥·z{Õ-¸û`‰TNu+o/DsŠ® ¶ÒÒ1°ÖNÿq>×f"…K,¾sP›´nÈ-sš$ 3¸½IWÂõ…_ŒDÉŒ½KRኻ8úãî ¬X딀ͱ@0:Íeäò+_¸œãè€zª{ûvæ_.áºcyµÞf!XÙ1åtdÅßñ:“G•fˆvSëóâ-y[2ìwq)ÌåK<¦=6aˆrÄh¥Yj\–…Ã]Ïzªû>Ôº=ö¢3yé6`@”PÆÁ.a†2D7'Ô ÐÅhGŸÙ£ì ׬¿‘[¹˜}© £[L‘;fï$ß.H¶ëxøKç)„&FÙÏX±2“aŒTj¨¼•H*%u&Údc9ëJyzwtÈÀ¼KnÝ'4ì-"ÿô²V±–34UÏöƒk+ |ß´Góá>åv¢Ý͉1†Ôj“pÍ’›Ž† >Eº„‘þô«hòÎ솾n¢à‡«Ï ¬ (rbº k6-–üâ@”&äí¤°‚ŸWF•À+¯W^Üã~ˆå„ü:*,®bÊ3È¡ÑLx (Èéç=Ø» É>Â%!ö6ÿÚä)ûñaãQD퇴ƒ¹a¼øIú2'’¿:µ£¢iývY ¨¶t|œxS”^[BJ9)~[Œ+ñÈ•‹w4tlúI±ùX‰°†IëgÈñƒâMâצxn˜g6ß?Wž»Q u£'mÌGG(éñ”¡1¾ óשÂr·¢£I‘æ&‹ ~îˆ1.–_¸aJÿžÉŠ!‚ ±„ÔáýWS†Y‡\PQ?WÀ"ï Ú CŠ!4v|³ß i&ã«u¨»ÀÃÏ2,1|úã ×±ËlÝx~[ò•lšÃ^®v˜¨‘Õzpê élGXòÆ7]ÚntvïZ„©p½Ù+~w5ËiZ1x~”d¸ÆÀÛÛþ÷VŠü?Q‡·eÙØÏæ¾bµCho‘¼ Eô“Ç×¥AÏuƒ±Q*_¬ÂJâ,§=- ©6_DQjÈ ˆrƉ³/µ€uAsD‘TÁxmúšÓÍšÞÚEdJ¾3!üþ_§ÌÄBô×j8!??0HÔ –~:µ¾?VÒ1Z]‰›j­AËÝuÀÇçT­/É.Dä¯s­Úo]‡Ó®3ÈÁÛ™´[æÐÒ_o)ñ:±çÏX•kûŒ¯F™MÇ8Øÿíãø”5¡„¸)#öD7!Ñv· \cÒ‡tM»5$’è)½þP=Ñö›ÉàéâO™Ÿè±°7W•«‚ ±Ípæ>N"`UyG³Óïs¾áåÝ=[·æ³–F„0õ›WëÃ÷÷O,´ ~Ú' •¡D¢pZ]ÇÖù»g£B1Ödm}ª¶óNNu*™cëtóI*צ: iï¸ìåÕ+a>¾zš-àIâw6³v "[ÁÉ©á0H°¨þƹ|m•éÎ JÄ ³Ögúu¨Î&ËTÚd½½  jœ¡˜‰³å‰üQ<“Çj/ÏA•yÅBg2FãjWžuÏ ìSb½ 3²Ä˜Þ±!¿±/ðë/>襒ÙH?4ìœLÁ˜(3ÄzºSÃí=‘¦‚`‚Þöás—ùéE¿GÞec4úzýIVþ#`Œ•šlã.Ún¿c>øÎ+âöz ½£ŽL¤3œ£ ÷)øp–¼—4¢¹ ôíœA¤x:zEB ˆ@d[kî£ 5½ÛÆfG¥æ@[tl¤JX’šµO;×âÓü#°áðNîG¹v 3¸› GGRÔ®ôíF¼Fgâèì²€²¦I«ÇpzÖðÉZ»Ì±¡”£”ç ñò2Ï_ÅÂ:ºýú yLð*‚Õ]ù;&I'.è:Ê&øC-v2ƒqH`x_:“.2f S^¿ š¢2Ù"ªô…üwíLˆ4±ßrQ –5ËØïû­Y“`ñ×=ñFîB#GD`]qïÖ“@²tíTbºê£¬” —›Ÿ¸§‚TRÌblÏf:„6©ý8gÉhíe{¡Å#ˆ"3Âñ}¬†e7©œ½b¶%ëC)Z~‹£’J¥îãOׇ¢ªw”Æà=1}襊ŒÆ´¥”nÑõ×)0(ÐæÔ$SÓjv~»T+õ«ÊÝv!]i±ò‹Bý Sž–¬RyºÄPf.Ý­{p( …$JZn(åæ_b}ÄùÆäÁàÊ[õ T8Ÿ¤‹3Û$ø.ïÞ&(×ɵx«s dDþûH¢¤5h½(ÚkÑã4c6Ú C›_Æ Å!€ŸR ÉàÞmM¡JL’+vO"¨ŸKþó Oæ;¶˜ã¸“Ôl¨ oöa!†)S•Ó¢Ú2¯¡Èn½´sYqGdJ·Se>k¦$\«Ù*p#ò1Ã/Χk¿¢±3siwÌŸ8:êzTúÙJÚ›"ŸŠ{¢Ê¦`r‡ ˆ/ñÁÛ&o¸h¥nn~ç&Š/ÀdÿsD8©Åm˜—Á_%ÂsD>}¬rFѲ<ÃäHÚQFQkÿ7VnÕ]ý:pÑ \’Fÿš=LÒ†ý%¹Ã ú5,¬»ÓÞuÇ]ßV`x±Ô¥rº½™ìdñüýç@±E½;?9J²[ä\©Õ1“Ú‰™ÆyÉË–Ðj]ìES¦?aØ–˜wÇÎmoÖ¥ KÏæ5(9¼»áìI„OøÀGlHð'ÝZ<£„B…u£ef¿¨š-hÈ›ñÇÅ—¤ŸF…ˆî¾æ¯ ›ÚB3Þy𠆟C·ôºGií·öŸôìÛËØÂH«]ñù"ü’†sF Š€Î…!Șö ¼€–õïsÑÛ[ ¡·•P¥HÑYm ÎÔòæx>[aQ磟­K,üªm¡E’LeË1÷´Pé§ñ(F<²­Ó{L±Æ-Û¿#Y…2Ç…W<%Ÿì‘+eÝÀ‘aŽX} ¼Œtsù ÁÆ ¡= Xä劻lJ—/‚ba/ ëK½ÇBGw×-µù­Ã”:É‚ÿøÀ=toàæ\@÷I 1æ‘òò›ÏßT|q>‹ÀÑšî½Cý_nº$z“ ˆgM-R@D>úÔ;ÿ¦·J³ °–øÁ°1„2]ƒ?DI›äãov€³³ØA7jñü{<€š¦ñT~âúSÆßQTx“ØÏ\r¨É´­YW÷ÐÒÇ©åJà$ÕYZ|ÐJ/‘1!·š†â𱯠±C‚K+‚…ýiÁ§‰ƒ“Ükût³ÈȇŪï ÒB¯E:ŽëÄH~NK½Á'ÓÕU)Í98!Ã=´•`³)}Çœ£6ÇgF·2{¿rÓÑ5\ùŸ]ÛD @ãI2…¡ð&ˆV<˜ã„&pN†Éà,SE9^ÖÉyýK6áÕ€ŽpœÈ0í…H10}¤µu7”Hj|å̜ޣ+|×jÜðR0sâÔÑÉ&^7™_êe´©ÍO2àP½y#ÇùS‚÷4o‰zûƒéÂöÙÿ*úh´?„Âé‡ùÂZ>]C{ýÜê·({Ê?‘µÝКØ*Z6ÀulmõåwN ()èVDöà~ìÓãk]09á3PÂî÷åM½3ÜßSMæ‚šmþ´84—ãžø)OXÖA¤+]dù%ëEëw8†k)È'%`Ûæ¡wQß| P4#õb—»Eì ¿wĹ=³³ ìp‰`´\¤T`t c\§—JF…ׄ·§Œ'& ìï](ÆZÐe-¢ê\ëJr½ðtåä]u1Óf™.eÀ÷$ç2AkàØn¡ÿÈÍÝ÷?õœ$cGñèž‚8"¶˜ªJ:ÕwØñ[çþÔ|åQà]+œ5ƒ¬"-Ön> ./»üéè^PÖN'…–2Ü0hyçå ÝQEŒyñ]¢˜PØLœ·%4s|6é66j¥½FHþò×Ïî=Õ©~îfn=ÔxߎUûq Rœœ:~ÓšÕÂ|´ ô”d¯Æ[€„Þ©¾:•ÙíÂNþX'ùŒ9rÕñ„qÌöm<þ¬Ô‰#v‰°›ÏË»zŸl"i"ŒE»$&†à;¾õ2CÐT¹!ûîVÍLO¸N{¹ GÉä"¼Û›©N{k®T[R ê\¯µ½:ñ±VݦŽ ~k˜š+8„‡9€,ı>Ú¤Ø)™8déÊÎnȽç:‹ød²ÖA´íÓU†LÄ*PQ/fœë$& ”jø ŸúxR¥9ÿâÍ6|0§ ¯­ô²ñI(q#aµô+ûûbxXÎ?¿Eu=/ð±Ú›Æ=ŒÖùlºÅ¿ìÓîO¸µW#ê¨O 7f“žû£`™új÷âV‘{êZ£Bâ^àxË„:Ru0èþý·/plà;ÿ7÷Y&žü”¾¼6Xò%n© ÉCè„Å˦“ØܯÆÚ”vñy"3‡o¤XûD/¡èÀ›æ""n Öí%­/JurÑ¿FZÑÿéóìÚaµÉ­¨^ rJkçu[ÿ°]!û9 iŽÀ ŠÜdÔ⡷д8:m²þ'a~M Þ`cŒÿ Z2¶BxO™“×uÍÉ ´Iq#”¡ÌVºS™úi™Àú6Í,ÕÜE !¦S)È^Ž¦OüãZÑÊÙ·¾ÙЮÌ ¦Z®m{î ;КÕþü…>¶Rç$‰D ùf¥ØÎÏU|ÿÚ´¦dé+œï€½=nÊLêÕH tÝÜ|ÈD™t¸Ç ¶2Ø/±1Ò¨ ¦ó= û¶ÇJ"lï¥%'Y•¥2 QZ»§{»pZ1‘Ž˜Ã‘“ôJò¨Œt­¬÷ô¯ƒ[f&‚Š%1¯*•}z½¼Ð›ç90tJúà®X^!] éŸVÐÓYËýxs2-·ýöhb««nC8¸=ž£FK 5V@³äó{¡mÕr£½²ê9~|Œj[ Õ½äÒ ª‹¦öÐðrÓNñÈçó¢H2%ü#œt¸0R É7!ðÒL[¥$©NP#õ³…!–†\†çÿ‘aZM:Ãͨ÷Ó²vÉ$âÄO:^¼—¹¡É`Jã®D’¿›LΑ$‡ä·¼-ÕKÑßoìHS˜$þÑ )˜Aõ— ‰¨j²mƒ²côGĈ3Æ ïH|8l[Wি(B²–yþÍ€ Âç:ñjõײoí=+ŒD¨ªwþèëè4}E¬P§k{ß sŸ•ÿ­¢€€*ø‹uîë†rz^Ë*ÂQAÓ}õ¹£„MúéYv]£0øAƒÃïÄÆšì4öÛžI öú ¶wÍ”3ªP5öÁµû=19¦Y×úJ:º{§¤Yw~ˆ<λz>´.ï 7ëyŽvS5±B“ÐNUxǵRæ}­÷BºŽôåpˆw¥YÃ’‰·²­Îe°E.Ï—ÆÚ¨BÅøiÛ²„e»JLþÖû²Xضöïß-® ŠF(ñœ~e®ËÇßýðä©°ºcÅä7à×dDùæåʼn¥4ºÀ·ðþîkðC/ÍTŽÒýt³ˆ^Ñï:b²+íª:J9YTq Õ ¬OMŸ ]ðhŸï!¤T­#-ô³D:êÂzóûèYLëG ÛKµƒm ¢Æù)Ð =u­ƒ#Ü÷ª™ *Ü ” >7Çé“€{±:‰Ðи–Œ ÏãG-Æ¿2‡½boñvÐÌýc"+($µŠç²b9cyÝuQŠ Èlƒ¸Ea‡bÐg»•,ƒ¹^'"_/º€?µíZWçÕ)Çga çpéµ6g?û_vÉ#oô^=TÃúm.K…†–)ÄXQSJvºmÂ?ßB4æÏ,”ãÇL͘.Z–uÈÕê*u@&øØÏz;m¾«‰ÍHïþé™DJ7´èã–&PÙžÐ# B‰jß,4¨¸p; m@_ !ŸùµÛTžoõ:ç•ü Þ‘¸ä>ßeùVN—|ð}‘Ùü"Ò¨à¹æ€‘*P󢚫Ë:âí!œ’ aà-zøÃú­ÉÞÈú‹ß»1n«BÆ#vÛ·>9Å©X~aL~ÿm¦¤ÏϱMÆRQ¦/.\B”ú¥º Õy¸¸^[_Oâì…µÓº/ê°}xWMx÷Û7È·aäEÇ$ÌÍò¥‚á 7³v]öФo#-u¾ÚÛÁ¹Ž¬iù=wÔX=tåBZI„ÎøtÅ5${:1s¼ÇŠ(4•.B3#î•Ëôj 7"*«×¡Æ¤ïU®dc65}³­ÝžẢQ'máÝÎa®“ Œ,Bô,?EV´èa¯¿ó™UÐM?s“DélÓo¬A…CX9ýŸ}ÄìCÆÅæçµrø>< _ƒÅpã&ÿøŽžë¾Õv¿öÿýBn1.?\±CÅþ笾{ôÂwUø´>¿@0U¥Åe•X@<[—t.Ω|©n}Úå•Ê3PaXûÞw‚½ã¾ÛJ4Xõº-JÊõ¡œžà:c_çïæáyã“ÊK˜¡\Ø-1âÚY\ܘ: ¹\Û [Fߨ—Þß¾*'å½]–ÈuƒÌ‹îùƒm×u“‘‡ ê½VØöÆ~|HÀø]Œ}y\¼@b_Bn¹@ߨ¡"á!¸/øn’š0ä³Ù¿j]ÜçÞÚý³½7•£G žÞ#ƒ(!„é—žBw RÓúì…c¾1ž^5ÙÓš¬]‡Õ!jfr€‡gz^+¥£R8AK¶G=Vî!Þ€l¸:¸ùmÁLLû´À8^B¶S·ì‰Å˜ì«;“°ö¹@Ì—'Ž³˜;ª?v7ÔîsÓÏìÈMÝtÂ#+zå½Á¯¼ð â:·HX‰)ßy¨ýÊ„—âäM3êØV+Ô­vµœádÛ¯Xžò‘**{xZhéÚ!«s›:ŠÈw—~“…î][¿æ^è‹îÒ@gQïl¹ÔÙro£ÄA 'fr¸ƒé)ÀmðÚÉ®<Ÿ~â´­ƒ‹Jq ‘xM9 endstream endobj 1368 0 obj << /Type /FontDescriptor /FontName /BIRPMW+Sabon-Italic /Flags 4 /FontBBox [-146 -228 1059 927] /Ascent 747 /CapHeight 706 /Descent -220 /ItalicAngle -12 /StemV 93 /XHeight 457 /CharSet (/A/B/C/E/F/G/I/L/M/N/P/S/V/a/b/c/d/e/f/fi/g/h/hyphen/i/k/l/m/n/o/p/period/quoteright/r/s/slash/t/u/v/w/x/y/z) /FontFile 1367 0 R >> endobj 1233 0 obj << /Type /Encoding /Differences [2/fi/fl 38/ampersand/quoteright/parenleft/parenright 44/comma/hyphen/period/slash/zero/one/two/three/four/five/six/seven/eight/nine/colon/semicolon 65/A/B/C/D/E/F/G/H/I/J/K/L/M/N/O/P 82/R/S/T/U/V/W/X 92/backslash 97/a/b/c/d/e/f/g/h/i/j/k/l/m/n/o/p/q/r/s/t/u/v/w/x/y/z 149/bullet 152/tilde 178/twosuperior 233/eacute] >> endobj 206 0 obj << /Type /Font /Subtype /Type1 /BaseFont /JTQSEL+CMEX10 /FontDescriptor 1336 0 R /FirstChar 0 /LastChar 113 /Widths 1162 0 R >> endobj 223 0 obj << /Type /Font /Subtype /Type1 /BaseFont /IGNLSA+CMSY10 /FontDescriptor 1338 0 R /FirstChar 1 /LastChar 112 /Widths 1155 0 R >> endobj 565 0 obj << /Type /Font /Subtype /Type1 /BaseFont /HWTUNG+CMSY7 /FontDescriptor 1340 0 R /FirstChar 3 /LastChar 112 /Widths 1154 0 R >> endobj 566 0 obj << /Type /Font /Subtype /Type1 /BaseFont /KJNCRM+EUEX10 /FontDescriptor 1342 0 R /FirstChar 14 /LastChar 125 /Widths 1153 0 R >> endobj 203 0 obj << /Type /Font /Subtype /Type1 /BaseFont /OATESS+EUFM10 /FontDescriptor 1344 0 R /FirstChar 40 /LastChar 93 /Widths 1164 0 R >> endobj 207 0 obj << /Type /Font /Subtype /Type1 /BaseFont /HSNTCT+EUFM7 /FontDescriptor 1346 0 R /FirstChar 40 /LastChar 61 /Widths 1161 0 R >> endobj 582 0 obj << /Type /Font /Subtype /Type1 /BaseFont /BYCRVB+EURB10 /FontDescriptor 1348 0 R /FirstChar 3 /LastChar 32 /Widths 1149 0 R >> endobj 581 0 obj << /Type /Font /Subtype /Type1 /BaseFont /BCJSZT+EURB7 /FontDescriptor 1350 0 R /FirstChar 3 /LastChar 32 /Widths 1150 0 R >> endobj 201 0 obj << /Type /Font /Subtype /Type1 /BaseFont /HKOQTE+EURM10 /FontDescriptor 1352 0 R /FirstChar 0 /LastChar 122 /Widths 1166 0 R >> endobj 202 0 obj << /Type /Font /Subtype /Type1 /BaseFont /SIDADR+EURM7 /FontDescriptor 1354 0 R /FirstChar 0 /LastChar 122 /Widths 1165 0 R >> endobj 204 0 obj << /Type /Font /Subtype /Type1 /BaseFont /EJLECR+EUSM10 /FontDescriptor 1356 0 R /FirstChar 68 /LastChar 106 /Widths 1163 0 R >> endobj 686 0 obj << /Type /Font /Subtype /Type1 /BaseFont /MCIEGV+EUSM7 /FontDescriptor 1358 0 R /FirstChar 68 /LastChar 106 /Widths 1148 0 R >> endobj 574 0 obj << /Type /Font /Subtype /Type1 /BaseFont /YFMQYV+MSAM10 /FontDescriptor 1360 0 R /FirstChar 62 /LastChar 88 /Widths 1152 0 R >> endobj 580 0 obj << /Type /Font /Subtype /Type1 /BaseFont /FITBVO+MSBM10 /FontDescriptor 1362 0 R /FirstChar 69 /LastChar 86 /Widths 1151 0 R >> endobj 148 0 obj << /Type /Font /Subtype /Type1 /BaseFont /OZLTXQ+Sabon-Bold /FontDescriptor 1364 0 R /FirstChar 39 /LastChar 121 /Widths 1326 0 R /Encoding 1233 0 R >> endobj 146 0 obj << /Type /Font /Subtype /Type1 /BaseFont /UIRQAT+Sabon-Roman /FontDescriptor 1366 0 R /FirstChar 2 /LastChar 233 /Widths 1334 0 R /Encoding 1233 0 R >> endobj 150 0 obj << /Type /Font /Subtype /Type1 /BaseFont /BIRPMW+Sabon-Italic /FontDescriptor 1368 0 R /FirstChar 2 /LastChar 122 /Widths 1234 0 R /Encoding 1233 0 R >> endobj 151 0 obj << /Type /Pages /Count 6 /Parent 1369 0 R /Kids [138 0 R 186 0 R 198 0 R 218 0 R 306 0 R 370 0 R] >> endobj 458 0 obj << /Type /Pages /Count 6 /Parent 1369 0 R /Kids [424 0 R 464 0 R 513 0 R 561 0 R 571 0 R 577 0 R] >> endobj 642 0 obj << /Type /Pages /Count 6 /Parent 1369 0 R /Kids [587 0 R 645 0 R 692 0 R 735 0 R 792 0 R 831 0 R] >> endobj 838 0 obj << /Type /Pages /Count 6 /Parent 1369 0 R /Kids [835 0 R 840 0 R 844 0 R 848 0 R 854 0 R 907 0 R] >> endobj 954 0 obj << /Type /Pages /Count 5 /Parent 1369 0 R /Kids [929 0 R 958 0 R 1006 0 R 1069 0 R 1087 0 R] >> endobj 1369 0 obj << /Type /Pages /Count 29 /Kids [151 0 R 458 0 R 642 0 R 838 0 R 954 0 R] >> endobj 1370 0 obj << /Type /Outlines /First 7 0 R /Last 123 0 R /Count 6 >> endobj 135 0 obj << /Title 136 0 R /A 133 0 R /Parent 123 0 R /Prev 131 0 R >> endobj 131 0 obj << /Title 132 0 R /A 129 0 R /Parent 123 0 R /Prev 127 0 R /Next 135 0 R >> endobj 127 0 obj << /Title 128 0 R /A 125 0 R /Parent 123 0 R /Next 131 0 R >> endobj 123 0 obj << /Title 124 0 R /A 121 0 R /Parent 1370 0 R /Prev 107 0 R /First 127 0 R /Last 135 0 R /Count -3 >> endobj 119 0 obj << /Title 120 0 R /A 117 0 R /Parent 107 0 R /Prev 115 0 R >> endobj 115 0 obj << /Title 116 0 R /A 113 0 R /Parent 107 0 R /Prev 111 0 R /Next 119 0 R >> endobj 111 0 obj << /Title 112 0 R /A 109 0 R /Parent 107 0 R /Next 115 0 R >> endobj 107 0 obj << /Title 108 0 R /A 105 0 R /Parent 1370 0 R /Prev 35 0 R /Next 123 0 R /First 111 0 R /Last 119 0 R /Count -3 >> endobj 103 0 obj << /Title 104 0 R /A 101 0 R /Parent 99 0 R >> endobj 99 0 obj << /Title 100 0 R /A 97 0 R /Parent 35 0 R /Prev 91 0 R /First 103 0 R /Last 103 0 R /Count -1 >> endobj 95 0 obj << /Title 96 0 R /A 93 0 R /Parent 91 0 R >> endobj 91 0 obj << /Title 92 0 R /A 89 0 R /Parent 35 0 R /Prev 75 0 R /Next 99 0 R /First 95 0 R /Last 95 0 R /Count -1 >> endobj 87 0 obj << /Title 88 0 R /A 85 0 R /Parent 75 0 R /Prev 83 0 R >> endobj 83 0 obj << /Title 84 0 R /A 81 0 R /Parent 75 0 R /Prev 79 0 R /Next 87 0 R >> endobj 79 0 obj << /Title 80 0 R /A 77 0 R /Parent 75 0 R /Next 83 0 R >> endobj 75 0 obj << /Title 76 0 R /A 73 0 R /Parent 35 0 R /Prev 55 0 R /Next 91 0 R /First 79 0 R /Last 87 0 R /Count -3 >> endobj 71 0 obj << /Title 72 0 R /A 69 0 R /Parent 55 0 R /Prev 67 0 R >> endobj 67 0 obj << /Title 68 0 R /A 65 0 R /Parent 55 0 R /Prev 63 0 R /Next 71 0 R >> endobj 63 0 obj << /Title 64 0 R /A 61 0 R /Parent 55 0 R /Prev 59 0 R /Next 67 0 R >> endobj 59 0 obj << /Title 60 0 R /A 57 0 R /Parent 55 0 R /Next 63 0 R >> endobj 55 0 obj << /Title 56 0 R /A 53 0 R /Parent 35 0 R /Prev 51 0 R /Next 75 0 R /First 59 0 R /Last 71 0 R /Count -4 >> endobj 51 0 obj << /Title 52 0 R /A 49 0 R /Parent 35 0 R /Prev 47 0 R /Next 55 0 R >> endobj 47 0 obj << /Title 48 0 R /A 45 0 R /Parent 35 0 R /Prev 43 0 R /Next 51 0 R >> endobj 43 0 obj << /Title 44 0 R /A 41 0 R /Parent 35 0 R /Prev 39 0 R /Next 47 0 R >> endobj 39 0 obj << /Title 40 0 R /A 37 0 R /Parent 35 0 R /Next 43 0 R >> endobj 35 0 obj << /Title 36 0 R /A 33 0 R /Parent 1370 0 R /Prev 15 0 R /Next 107 0 R /First 39 0 R /Last 99 0 R /Count -8 >> endobj 31 0 obj << /Title 32 0 R /A 29 0 R /Parent 15 0 R /Prev 27 0 R >> endobj 27 0 obj << /Title 28 0 R /A 25 0 R /Parent 15 0 R /Prev 23 0 R /Next 31 0 R >> endobj 23 0 obj << /Title 24 0 R /A 21 0 R /Parent 15 0 R /Prev 19 0 R /Next 27 0 R >> endobj 19 0 obj << /Title 20 0 R /A 17 0 R /Parent 15 0 R /Next 23 0 R >> endobj 15 0 obj << /Title 16 0 R /A 13 0 R /Parent 1370 0 R /Prev 11 0 R /Next 35 0 R /First 19 0 R /Last 31 0 R /Count -4 >> endobj 11 0 obj << /Title 12 0 R /A 9 0 R /Parent 1370 0 R /Prev 7 0 R /Next 15 0 R >> endobj 7 0 obj << /Title 8 0 R /A 5 0 R /Parent 1370 0 R /Next 11 0 R >> endobj 1371 0 obj << /Names [(Doc-Start) 145 0 R (equation.1.1) 205 0 R (equation.3.2) 564 0 R (equation.4.3) 583 0 R (lstlisting.-1) 224 0 R (lstlisting.-10) 441 0 R] /Limits [(Doc-Start) (lstlisting.-10)] >> endobj 1372 0 obj << /Names [(lstlisting.-11) 467 0 R (lstlisting.-12) 516 0 R (lstlisting.-13) 590 0 R (lstlisting.-14) 695 0 R (lstlisting.-15) 746 0 R (lstlisting.-16) 763 0 R] /Limits [(lstlisting.-11) (lstlisting.-16)] >> endobj 1373 0 obj << /Names [(lstlisting.-17) 795 0 R (lstlisting.-18) 810 0 R (lstlisting.-19) 857 0 R (lstlisting.-2) 235 0 R (lstlisting.-20) 913 0 R (lstlisting.-21) 961 0 R] /Limits [(lstlisting.-17) (lstlisting.-21)] >> endobj 1374 0 obj << /Names [(lstlisting.-22) 1090 0 R (lstlisting.-3) 316 0 R (lstlisting.-4) 318 0 R (lstlisting.-5) 323 0 R (lstlisting.-6) 331 0 R (lstlisting.-7) 348 0 R] /Limits [(lstlisting.-22) (lstlisting.-7)] >> endobj 1375 0 obj << /Names [(lstlisting.-8) 379 0 R (lstlisting.-9) 406 0 R (lstnumber.-1.1) 225 0 R (lstnumber.-1.2) 226 0 R (lstnumber.-1.3) 227 0 R (lstnumber.-1.4) 228 0 R] /Limits [(lstlisting.-8) (lstnumber.-1.4)] >> endobj 1376 0 obj << /Names [(lstnumber.-1.5) 229 0 R (lstnumber.-1.6) 230 0 R (lstnumber.-1.7) 231 0 R (lstnumber.-1.8) 232 0 R (lstnumber.-1.9) 233 0 R (lstnumber.-10.1) 442 0 R] /Limits [(lstnumber.-1.5) (lstnumber.-10.1)] >> endobj 1377 0 obj << /Names [(lstnumber.-10.10) 451 0 R (lstnumber.-10.11) 452 0 R (lstnumber.-10.12) 453 0 R (lstnumber.-10.13) 454 0 R (lstnumber.-10.14) 455 0 R (lstnumber.-10.15) 456 0 R] /Limits [(lstnumber.-10.10) (lstnumber.-10.15)] >> endobj 1378 0 obj << /Names [(lstnumber.-10.16) 457 0 R (lstnumber.-10.2) 443 0 R (lstnumber.-10.3) 444 0 R (lstnumber.-10.4) 445 0 R (lstnumber.-10.5) 446 0 R (lstnumber.-10.6) 447 0 R] /Limits [(lstnumber.-10.16) (lstnumber.-10.6)] >> endobj 1379 0 obj << /Names [(lstnumber.-10.7) 448 0 R (lstnumber.-10.8) 449 0 R (lstnumber.-10.9) 450 0 R (lstnumber.-11.1) 468 0 R (lstnumber.-11.10) 477 0 R (lstnumber.-11.11) 478 0 R] /Limits [(lstnumber.-10.7) (lstnumber.-11.11)] >> endobj 1380 0 obj << /Names [(lstnumber.-11.12) 479 0 R (lstnumber.-11.13) 480 0 R (lstnumber.-11.14) 481 0 R (lstnumber.-11.15) 482 0 R (lstnumber.-11.16) 483 0 R (lstnumber.-11.17) 484 0 R] /Limits [(lstnumber.-11.12) (lstnumber.-11.17)] >> endobj 1381 0 obj << /Names [(lstnumber.-11.18) 485 0 R (lstnumber.-11.19) 486 0 R (lstnumber.-11.2) 469 0 R (lstnumber.-11.20) 487 0 R (lstnumber.-11.21) 488 0 R (lstnumber.-11.22) 489 0 R] /Limits [(lstnumber.-11.18) (lstnumber.-11.22)] >> endobj 1382 0 obj << /Names [(lstnumber.-11.23) 490 0 R (lstnumber.-11.24) 491 0 R (lstnumber.-11.25) 492 0 R (lstnumber.-11.26) 493 0 R (lstnumber.-11.27) 494 0 R (lstnumber.-11.28) 495 0 R] /Limits [(lstnumber.-11.23) (lstnumber.-11.28)] >> endobj 1383 0 obj << /Names [(lstnumber.-11.29) 496 0 R (lstnumber.-11.3) 470 0 R (lstnumber.-11.30) 497 0 R (lstnumber.-11.31) 498 0 R (lstnumber.-11.32) 499 0 R (lstnumber.-11.33) 500 0 R] /Limits [(lstnumber.-11.29) (lstnumber.-11.33)] >> endobj 1384 0 obj << /Names [(lstnumber.-11.34) 501 0 R (lstnumber.-11.35) 502 0 R (lstnumber.-11.36) 503 0 R (lstnumber.-11.37) 504 0 R (lstnumber.-11.38) 505 0 R (lstnumber.-11.39) 506 0 R] /Limits [(lstnumber.-11.34) (lstnumber.-11.39)] >> endobj 1385 0 obj << /Names [(lstnumber.-11.4) 471 0 R (lstnumber.-11.40) 507 0 R (lstnumber.-11.41) 508 0 R (lstnumber.-11.5) 472 0 R (lstnumber.-11.6) 473 0 R (lstnumber.-11.7) 474 0 R] /Limits [(lstnumber.-11.4) (lstnumber.-11.7)] >> endobj 1386 0 obj << /Names [(lstnumber.-11.8) 475 0 R (lstnumber.-11.9) 476 0 R (lstnumber.-12.1) 517 0 R (lstnumber.-12.10) 526 0 R (lstnumber.-12.11) 527 0 R (lstnumber.-12.12) 528 0 R] /Limits [(lstnumber.-11.8) (lstnumber.-12.12)] >> endobj 1387 0 obj << /Names [(lstnumber.-12.13) 529 0 R (lstnumber.-12.14) 530 0 R (lstnumber.-12.15) 531 0 R (lstnumber.-12.16) 532 0 R (lstnumber.-12.17) 533 0 R (lstnumber.-12.18) 534 0 R] /Limits [(lstnumber.-12.13) (lstnumber.-12.18)] >> endobj 1388 0 obj << /Names [(lstnumber.-12.19) 535 0 R (lstnumber.-12.2) 518 0 R (lstnumber.-12.20) 536 0 R (lstnumber.-12.21) 537 0 R (lstnumber.-12.22) 538 0 R (lstnumber.-12.23) 539 0 R] /Limits [(lstnumber.-12.19) (lstnumber.-12.23)] >> endobj 1389 0 obj << /Names [(lstnumber.-12.24) 540 0 R (lstnumber.-12.25) 541 0 R (lstnumber.-12.26) 542 0 R (lstnumber.-12.27) 543 0 R (lstnumber.-12.28) 544 0 R (lstnumber.-12.29) 545 0 R] /Limits [(lstnumber.-12.24) (lstnumber.-12.29)] >> endobj 1390 0 obj << /Names [(lstnumber.-12.3) 519 0 R (lstnumber.-12.30) 546 0 R (lstnumber.-12.31) 547 0 R (lstnumber.-12.32) 548 0 R (lstnumber.-12.33) 549 0 R (lstnumber.-12.34) 550 0 R] /Limits [(lstnumber.-12.3) (lstnumber.-12.34)] >> endobj 1391 0 obj << /Names [(lstnumber.-12.35) 551 0 R (lstnumber.-12.36) 552 0 R (lstnumber.-12.37) 553 0 R (lstnumber.-12.38) 554 0 R (lstnumber.-12.39) 555 0 R (lstnumber.-12.4) 520 0 R] /Limits [(lstnumber.-12.35) (lstnumber.-12.4)] >> endobj 1392 0 obj << /Names [(lstnumber.-12.40) 556 0 R (lstnumber.-12.41) 557 0 R (lstnumber.-12.5) 521 0 R (lstnumber.-12.6) 522 0 R (lstnumber.-12.7) 523 0 R (lstnumber.-12.8) 524 0 R] /Limits [(lstnumber.-12.40) (lstnumber.-12.8)] >> endobj 1393 0 obj << /Names [(lstnumber.-12.9) 525 0 R (lstnumber.-13.1) 591 0 R (lstnumber.-13.10) 600 0 R (lstnumber.-13.11) 601 0 R (lstnumber.-13.12) 602 0 R (lstnumber.-13.13) 603 0 R] /Limits [(lstnumber.-12.9) (lstnumber.-13.13)] >> endobj 1394 0 obj << /Names [(lstnumber.-13.14) 604 0 R (lstnumber.-13.15) 605 0 R (lstnumber.-13.16) 606 0 R (lstnumber.-13.17) 607 0 R (lstnumber.-13.18) 608 0 R (lstnumber.-13.19) 609 0 R] /Limits [(lstnumber.-13.14) (lstnumber.-13.19)] >> endobj 1395 0 obj << /Names [(lstnumber.-13.2) 592 0 R (lstnumber.-13.20) 610 0 R (lstnumber.-13.21) 611 0 R (lstnumber.-13.22) 612 0 R (lstnumber.-13.23) 613 0 R (lstnumber.-13.24) 614 0 R] /Limits [(lstnumber.-13.2) (lstnumber.-13.24)] >> endobj 1396 0 obj << /Names [(lstnumber.-13.25) 615 0 R (lstnumber.-13.26) 616 0 R (lstnumber.-13.27) 617 0 R (lstnumber.-13.28) 618 0 R (lstnumber.-13.29) 619 0 R (lstnumber.-13.3) 593 0 R] /Limits [(lstnumber.-13.25) (lstnumber.-13.3)] >> endobj 1397 0 obj << /Names [(lstnumber.-13.30) 620 0 R (lstnumber.-13.31) 621 0 R (lstnumber.-13.32) 622 0 R (lstnumber.-13.33) 623 0 R (lstnumber.-13.34) 624 0 R (lstnumber.-13.35) 625 0 R] /Limits [(lstnumber.-13.30) (lstnumber.-13.35)] >> endobj 1398 0 obj << /Names [(lstnumber.-13.36) 626 0 R (lstnumber.-13.37) 627 0 R (lstnumber.-13.38) 628 0 R (lstnumber.-13.39) 629 0 R (lstnumber.-13.4) 594 0 R (lstnumber.-13.40) 630 0 R] /Limits [(lstnumber.-13.36) (lstnumber.-13.40)] >> endobj 1399 0 obj << /Names [(lstnumber.-13.41) 631 0 R (lstnumber.-13.42) 632 0 R (lstnumber.-13.43) 633 0 R (lstnumber.-13.44) 634 0 R (lstnumber.-13.45) 635 0 R (lstnumber.-13.46) 636 0 R] /Limits [(lstnumber.-13.41) (lstnumber.-13.46)] >> endobj 1400 0 obj << /Names [(lstnumber.-13.47) 637 0 R (lstnumber.-13.48) 638 0 R (lstnumber.-13.49) 639 0 R (lstnumber.-13.5) 595 0 R (lstnumber.-13.50) 640 0 R (lstnumber.-13.51) 641 0 R] /Limits [(lstnumber.-13.47) (lstnumber.-13.51)] >> endobj 1401 0 obj << /Names [(lstnumber.-13.52) 648 0 R (lstnumber.-13.53) 649 0 R (lstnumber.-13.54) 650 0 R (lstnumber.-13.55) 651 0 R (lstnumber.-13.56) 652 0 R (lstnumber.-13.57) 653 0 R] /Limits [(lstnumber.-13.52) (lstnumber.-13.57)] >> endobj 1402 0 obj << /Names [(lstnumber.-13.58) 654 0 R (lstnumber.-13.59) 655 0 R (lstnumber.-13.6) 596 0 R (lstnumber.-13.60) 656 0 R (lstnumber.-13.61) 657 0 R (lstnumber.-13.62) 658 0 R] /Limits [(lstnumber.-13.58) (lstnumber.-13.62)] >> endobj 1403 0 obj << /Names [(lstnumber.-13.63) 659 0 R (lstnumber.-13.64) 660 0 R (lstnumber.-13.65) 661 0 R (lstnumber.-13.66) 662 0 R (lstnumber.-13.67) 663 0 R (lstnumber.-13.68) 664 0 R] /Limits [(lstnumber.-13.63) (lstnumber.-13.68)] >> endobj 1404 0 obj << /Names [(lstnumber.-13.69) 665 0 R (lstnumber.-13.7) 597 0 R (lstnumber.-13.70) 666 0 R (lstnumber.-13.71) 667 0 R (lstnumber.-13.72) 668 0 R (lstnumber.-13.73) 669 0 R] /Limits [(lstnumber.-13.69) (lstnumber.-13.73)] >> endobj 1405 0 obj << /Names [(lstnumber.-13.74) 670 0 R (lstnumber.-13.75) 671 0 R (lstnumber.-13.76) 672 0 R (lstnumber.-13.77) 673 0 R (lstnumber.-13.78) 674 0 R (lstnumber.-13.79) 675 0 R] /Limits [(lstnumber.-13.74) (lstnumber.-13.79)] >> endobj 1406 0 obj << /Names [(lstnumber.-13.8) 598 0 R (lstnumber.-13.80) 676 0 R (lstnumber.-13.81) 677 0 R (lstnumber.-13.82) 678 0 R (lstnumber.-13.83) 679 0 R (lstnumber.-13.84) 680 0 R] /Limits [(lstnumber.-13.8) (lstnumber.-13.84)] >> endobj 1407 0 obj << /Names [(lstnumber.-13.85) 681 0 R (lstnumber.-13.86) 682 0 R (lstnumber.-13.87) 683 0 R (lstnumber.-13.88) 684 0 R (lstnumber.-13.89) 685 0 R (lstnumber.-13.9) 599 0 R] /Limits [(lstnumber.-13.85) (lstnumber.-13.9)] >> endobj 1408 0 obj << /Names [(lstnumber.-14.1) 696 0 R (lstnumber.-14.10) 705 0 R (lstnumber.-14.11) 706 0 R (lstnumber.-14.12) 707 0 R (lstnumber.-14.13) 708 0 R (lstnumber.-14.14) 709 0 R] /Limits [(lstnumber.-14.1) (lstnumber.-14.14)] >> endobj 1409 0 obj << /Names [(lstnumber.-14.15) 710 0 R (lstnumber.-14.16) 711 0 R (lstnumber.-14.17) 712 0 R (lstnumber.-14.18) 713 0 R (lstnumber.-14.19) 714 0 R (lstnumber.-14.2) 697 0 R] /Limits [(lstnumber.-14.15) (lstnumber.-14.2)] >> endobj 1410 0 obj << /Names [(lstnumber.-14.20) 715 0 R (lstnumber.-14.21) 716 0 R (lstnumber.-14.22) 717 0 R (lstnumber.-14.23) 718 0 R (lstnumber.-14.24) 719 0 R (lstnumber.-14.25) 720 0 R] /Limits [(lstnumber.-14.20) (lstnumber.-14.25)] >> endobj 1411 0 obj << /Names [(lstnumber.-14.26) 721 0 R (lstnumber.-14.27) 722 0 R (lstnumber.-14.28) 738 0 R (lstnumber.-14.29) 739 0 R (lstnumber.-14.3) 698 0 R (lstnumber.-14.30) 740 0 R] /Limits [(lstnumber.-14.26) (lstnumber.-14.30)] >> endobj 1412 0 obj << /Names [(lstnumber.-14.31) 741 0 R (lstnumber.-14.32) 742 0 R (lstnumber.-14.33) 743 0 R (lstnumber.-14.34) 744 0 R (lstnumber.-14.35) 745 0 R (lstnumber.-14.4) 699 0 R] /Limits [(lstnumber.-14.31) (lstnumber.-14.4)] >> endobj 1413 0 obj << /Names [(lstnumber.-14.5) 700 0 R (lstnumber.-14.6) 701 0 R (lstnumber.-14.7) 702 0 R (lstnumber.-14.8) 703 0 R (lstnumber.-14.9) 704 0 R (lstnumber.-15.1) 747 0 R] /Limits [(lstnumber.-14.5) (lstnumber.-15.1)] >> endobj 1414 0 obj << /Names [(lstnumber.-15.10) 756 0 R (lstnumber.-15.11) 757 0 R (lstnumber.-15.12) 758 0 R (lstnumber.-15.13) 759 0 R (lstnumber.-15.14) 760 0 R (lstnumber.-15.15) 761 0 R] /Limits [(lstnumber.-15.10) (lstnumber.-15.15)] >> endobj 1415 0 obj << /Names [(lstnumber.-15.2) 748 0 R (lstnumber.-15.3) 749 0 R (lstnumber.-15.4) 750 0 R (lstnumber.-15.5) 751 0 R (lstnumber.-15.6) 752 0 R (lstnumber.-15.7) 753 0 R] /Limits [(lstnumber.-15.2) (lstnumber.-15.7)] >> endobj 1416 0 obj << /Names [(lstnumber.-15.8) 754 0 R (lstnumber.-15.9) 755 0 R (lstnumber.-16.1) 764 0 R (lstnumber.-16.10) 773 0 R (lstnumber.-16.11) 774 0 R (lstnumber.-16.12) 775 0 R] /Limits [(lstnumber.-15.8) (lstnumber.-16.12)] >> endobj 1417 0 obj << /Names [(lstnumber.-16.13) 776 0 R (lstnumber.-16.14) 777 0 R (lstnumber.-16.15) 778 0 R (lstnumber.-16.16) 779 0 R (lstnumber.-16.17) 780 0 R (lstnumber.-16.18) 781 0 R] /Limits [(lstnumber.-16.13) (lstnumber.-16.18)] >> endobj 1418 0 obj << /Names [(lstnumber.-16.19) 782 0 R (lstnumber.-16.2) 765 0 R (lstnumber.-16.3) 766 0 R (lstnumber.-16.4) 767 0 R (lstnumber.-16.5) 768 0 R (lstnumber.-16.6) 769 0 R] /Limits [(lstnumber.-16.19) (lstnumber.-16.6)] >> endobj 1419 0 obj << /Names [(lstnumber.-16.7) 770 0 R (lstnumber.-16.8) 771 0 R (lstnumber.-16.9) 772 0 R (lstnumber.-17.1) 796 0 R (lstnumber.-17.10) 805 0 R (lstnumber.-17.11) 806 0 R] /Limits [(lstnumber.-16.7) (lstnumber.-17.11)] >> endobj 1420 0 obj << /Names [(lstnumber.-17.12) 807 0 R (lstnumber.-17.13) 808 0 R (lstnumber.-17.14) 809 0 R (lstnumber.-17.2) 797 0 R (lstnumber.-17.3) 798 0 R (lstnumber.-17.4) 799 0 R] /Limits [(lstnumber.-17.12) (lstnumber.-17.4)] >> endobj 1421 0 obj << /Names [(lstnumber.-17.5) 800 0 R (lstnumber.-17.6) 801 0 R (lstnumber.-17.7) 802 0 R (lstnumber.-17.8) 803 0 R (lstnumber.-17.9) 804 0 R (lstnumber.-18.1) 811 0 R] /Limits [(lstnumber.-17.5) (lstnumber.-18.1)] >> endobj 1422 0 obj << /Names [(lstnumber.-18.10) 820 0 R (lstnumber.-18.11) 821 0 R (lstnumber.-18.12) 822 0 R (lstnumber.-18.13) 823 0 R (lstnumber.-18.14) 824 0 R (lstnumber.-18.15) 825 0 R] /Limits [(lstnumber.-18.10) (lstnumber.-18.15)] >> endobj 1423 0 obj << /Names [(lstnumber.-18.16) 826 0 R (lstnumber.-18.17) 827 0 R (lstnumber.-18.18) 828 0 R (lstnumber.-18.19) 829 0 R (lstnumber.-18.2) 812 0 R (lstnumber.-18.3) 813 0 R] /Limits [(lstnumber.-18.16) (lstnumber.-18.3)] >> endobj 1424 0 obj << /Names [(lstnumber.-18.4) 814 0 R (lstnumber.-18.5) 815 0 R (lstnumber.-18.6) 816 0 R (lstnumber.-18.7) 817 0 R (lstnumber.-18.8) 818 0 R (lstnumber.-18.9) 819 0 R] /Limits [(lstnumber.-18.4) (lstnumber.-18.9)] >> endobj 1425 0 obj << /Names [(lstnumber.-19.1) 858 0 R (lstnumber.-19.10) 867 0 R (lstnumber.-19.11) 868 0 R (lstnumber.-19.12) 869 0 R (lstnumber.-19.13) 870 0 R (lstnumber.-19.14) 871 0 R] /Limits [(lstnumber.-19.1) (lstnumber.-19.14)] >> endobj 1426 0 obj << /Names [(lstnumber.-19.15) 872 0 R (lstnumber.-19.16) 873 0 R (lstnumber.-19.17) 874 0 R (lstnumber.-19.18) 875 0 R (lstnumber.-19.19) 876 0 R (lstnumber.-19.2) 859 0 R] /Limits [(lstnumber.-19.15) (lstnumber.-19.2)] >> endobj 1427 0 obj << /Names [(lstnumber.-19.20) 877 0 R (lstnumber.-19.21) 878 0 R (lstnumber.-19.22) 879 0 R (lstnumber.-19.23) 880 0 R (lstnumber.-19.24) 881 0 R (lstnumber.-19.25) 882 0 R] /Limits [(lstnumber.-19.20) (lstnumber.-19.25)] >> endobj 1428 0 obj << /Names [(lstnumber.-19.26) 883 0 R (lstnumber.-19.27) 884 0 R (lstnumber.-19.28) 885 0 R (lstnumber.-19.29) 886 0 R (lstnumber.-19.3) 860 0 R (lstnumber.-19.30) 887 0 R] /Limits [(lstnumber.-19.26) (lstnumber.-19.30)] >> endobj 1429 0 obj << /Names [(lstnumber.-19.31) 888 0 R (lstnumber.-19.32) 889 0 R (lstnumber.-19.33) 890 0 R (lstnumber.-19.34) 891 0 R (lstnumber.-19.35) 892 0 R (lstnumber.-19.36) 893 0 R] /Limits [(lstnumber.-19.31) (lstnumber.-19.36)] >> endobj 1430 0 obj << /Names [(lstnumber.-19.37) 894 0 R (lstnumber.-19.38) 895 0 R (lstnumber.-19.39) 896 0 R (lstnumber.-19.4) 861 0 R (lstnumber.-19.40) 897 0 R (lstnumber.-19.41) 898 0 R] /Limits [(lstnumber.-19.37) (lstnumber.-19.41)] >> endobj 1431 0 obj << /Names [(lstnumber.-19.42) 899 0 R (lstnumber.-19.43) 900 0 R (lstnumber.-19.44) 901 0 R (lstnumber.-19.45) 910 0 R (lstnumber.-19.46) 911 0 R (lstnumber.-19.47) 912 0 R] /Limits [(lstnumber.-19.42) (lstnumber.-19.47)] >> endobj 1432 0 obj << /Names [(lstnumber.-19.5) 862 0 R (lstnumber.-19.6) 863 0 R (lstnumber.-19.7) 864 0 R (lstnumber.-19.8) 865 0 R (lstnumber.-19.9) 866 0 R (lstnumber.-2.1) 236 0 R] /Limits [(lstnumber.-19.5) (lstnumber.-2.1)] >> endobj 1433 0 obj << /Names [(lstnumber.-2.10) 245 0 R (lstnumber.-2.11) 246 0 R (lstnumber.-2.12) 247 0 R (lstnumber.-2.13) 248 0 R (lstnumber.-2.14) 249 0 R (lstnumber.-2.15) 250 0 R] /Limits [(lstnumber.-2.10) (lstnumber.-2.15)] >> endobj 1434 0 obj << /Names [(lstnumber.-2.16) 251 0 R (lstnumber.-2.17) 252 0 R (lstnumber.-2.18) 253 0 R (lstnumber.-2.19) 254 0 R (lstnumber.-2.2) 237 0 R (lstnumber.-2.20) 255 0 R] /Limits [(lstnumber.-2.16) (lstnumber.-2.20)] >> endobj 1435 0 obj << /Names [(lstnumber.-2.21) 256 0 R (lstnumber.-2.22) 257 0 R (lstnumber.-2.23) 258 0 R (lstnumber.-2.24) 259 0 R (lstnumber.-2.25) 260 0 R (lstnumber.-2.26) 261 0 R] /Limits [(lstnumber.-2.21) (lstnumber.-2.26)] >> endobj 1436 0 obj << /Names [(lstnumber.-2.27) 262 0 R (lstnumber.-2.28) 263 0 R (lstnumber.-2.29) 264 0 R (lstnumber.-2.3) 238 0 R (lstnumber.-2.30) 265 0 R (lstnumber.-2.31) 266 0 R] /Limits [(lstnumber.-2.27) (lstnumber.-2.31)] >> endobj 1437 0 obj << /Names [(lstnumber.-2.32) 267 0 R (lstnumber.-2.33) 268 0 R (lstnumber.-2.34) 269 0 R (lstnumber.-2.35) 270 0 R (lstnumber.-2.36) 271 0 R (lstnumber.-2.37) 272 0 R] /Limits [(lstnumber.-2.32) (lstnumber.-2.37)] >> endobj 1438 0 obj << /Names [(lstnumber.-2.38) 273 0 R (lstnumber.-2.39) 274 0 R (lstnumber.-2.4) 239 0 R (lstnumber.-2.40) 309 0 R (lstnumber.-2.41) 310 0 R (lstnumber.-2.42) 311 0 R] /Limits [(lstnumber.-2.38) (lstnumber.-2.42)] >> endobj 1439 0 obj << /Names [(lstnumber.-2.43) 312 0 R (lstnumber.-2.44) 313 0 R (lstnumber.-2.45) 314 0 R (lstnumber.-2.5) 240 0 R (lstnumber.-2.6) 241 0 R (lstnumber.-2.7) 242 0 R] /Limits [(lstnumber.-2.43) (lstnumber.-2.7)] >> endobj 1440 0 obj << /Names [(lstnumber.-2.8) 243 0 R (lstnumber.-2.9) 244 0 R (lstnumber.-20.1) 914 0 R (lstnumber.-20.10) 923 0 R (lstnumber.-20.11) 924 0 R (lstnumber.-20.12) 925 0 R] /Limits [(lstnumber.-2.8) (lstnumber.-20.12)] >> endobj 1441 0 obj << /Names [(lstnumber.-20.13) 926 0 R (lstnumber.-20.14) 927 0 R (lstnumber.-20.15) 932 0 R (lstnumber.-20.16) 933 0 R (lstnumber.-20.17) 934 0 R (lstnumber.-20.18) 935 0 R] /Limits [(lstnumber.-20.13) (lstnumber.-20.18)] >> endobj 1442 0 obj << /Names [(lstnumber.-20.19) 936 0 R (lstnumber.-20.2) 915 0 R (lstnumber.-20.20) 937 0 R (lstnumber.-20.21) 938 0 R (lstnumber.-20.22) 939 0 R (lstnumber.-20.23) 940 0 R] /Limits [(lstnumber.-20.19) (lstnumber.-20.23)] >> endobj 1443 0 obj << /Names [(lstnumber.-20.24) 941 0 R (lstnumber.-20.25) 942 0 R (lstnumber.-20.26) 943 0 R (lstnumber.-20.27) 944 0 R (lstnumber.-20.28) 945 0 R (lstnumber.-20.29) 946 0 R] /Limits [(lstnumber.-20.24) (lstnumber.-20.29)] >> endobj 1444 0 obj << /Names [(lstnumber.-20.3) 916 0 R (lstnumber.-20.30) 947 0 R (lstnumber.-20.31) 948 0 R (lstnumber.-20.32) 949 0 R (lstnumber.-20.33) 950 0 R (lstnumber.-20.34) 951 0 R] /Limits [(lstnumber.-20.3) (lstnumber.-20.34)] >> endobj 1445 0 obj << /Names [(lstnumber.-20.35) 952 0 R (lstnumber.-20.36) 953 0 R (lstnumber.-20.4) 917 0 R (lstnumber.-20.5) 918 0 R (lstnumber.-20.6) 919 0 R (lstnumber.-20.7) 920 0 R] /Limits [(lstnumber.-20.35) (lstnumber.-20.7)] >> endobj 1446 0 obj << /Names [(lstnumber.-20.8) 921 0 R (lstnumber.-20.9) 922 0 R (lstnumber.-21.1) 962 0 R (lstnumber.-21.10) 971 0 R (lstnumber.-21.100) 1065 0 R (lstnumber.-21.101) 1066 0 R] /Limits [(lstnumber.-20.8) (lstnumber.-21.101)] >> endobj 1447 0 obj << /Names [(lstnumber.-21.102) 1072 0 R (lstnumber.-21.103) 1073 0 R (lstnumber.-21.104) 1074 0 R (lstnumber.-21.105) 1075 0 R (lstnumber.-21.106) 1076 0 R (lstnumber.-21.107) 1077 0 R] /Limits [(lstnumber.-21.102) (lstnumber.-21.107)] >> endobj 1448 0 obj << /Names [(lstnumber.-21.108) 1078 0 R (lstnumber.-21.109) 1079 0 R (lstnumber.-21.11) 972 0 R (lstnumber.-21.110) 1080 0 R (lstnumber.-21.12) 973 0 R (lstnumber.-21.13) 974 0 R] /Limits [(lstnumber.-21.108) (lstnumber.-21.13)] >> endobj 1449 0 obj << /Names [(lstnumber.-21.14) 975 0 R (lstnumber.-21.15) 976 0 R (lstnumber.-21.16) 977 0 R (lstnumber.-21.17) 978 0 R (lstnumber.-21.18) 979 0 R (lstnumber.-21.19) 980 0 R] /Limits [(lstnumber.-21.14) (lstnumber.-21.19)] >> endobj 1450 0 obj << /Names [(lstnumber.-21.2) 963 0 R (lstnumber.-21.20) 981 0 R (lstnumber.-21.21) 982 0 R (lstnumber.-21.22) 983 0 R (lstnumber.-21.23) 984 0 R (lstnumber.-21.24) 985 0 R] /Limits [(lstnumber.-21.2) (lstnumber.-21.24)] >> endobj 1451 0 obj << /Names [(lstnumber.-21.25) 986 0 R (lstnumber.-21.26) 987 0 R (lstnumber.-21.27) 988 0 R (lstnumber.-21.28) 989 0 R (lstnumber.-21.29) 990 0 R (lstnumber.-21.3) 964 0 R] /Limits [(lstnumber.-21.25) (lstnumber.-21.3)] >> endobj 1452 0 obj << /Names [(lstnumber.-21.30) 991 0 R (lstnumber.-21.31) 992 0 R (lstnumber.-21.32) 993 0 R (lstnumber.-21.33) 994 0 R (lstnumber.-21.34) 995 0 R (lstnumber.-21.35) 996 0 R] /Limits [(lstnumber.-21.30) (lstnumber.-21.35)] >> endobj 1453 0 obj << /Names [(lstnumber.-21.36) 997 0 R (lstnumber.-21.37) 998 0 R (lstnumber.-21.38) 999 0 R (lstnumber.-21.39) 1000 0 R (lstnumber.-21.4) 965 0 R (lstnumber.-21.40) 1001 0 R] /Limits [(lstnumber.-21.36) (lstnumber.-21.40)] >> endobj 1454 0 obj << /Names [(lstnumber.-21.41) 1002 0 R (lstnumber.-21.42) 1003 0 R (lstnumber.-21.43) 1004 0 R (lstnumber.-21.44) 1009 0 R (lstnumber.-21.45) 1010 0 R (lstnumber.-21.46) 1011 0 R] /Limits [(lstnumber.-21.41) (lstnumber.-21.46)] >> endobj 1455 0 obj << /Names [(lstnumber.-21.47) 1012 0 R (lstnumber.-21.48) 1013 0 R (lstnumber.-21.49) 1014 0 R (lstnumber.-21.5) 966 0 R (lstnumber.-21.50) 1015 0 R (lstnumber.-21.51) 1016 0 R] /Limits [(lstnumber.-21.47) (lstnumber.-21.51)] >> endobj 1456 0 obj << /Names [(lstnumber.-21.52) 1017 0 R (lstnumber.-21.53) 1018 0 R (lstnumber.-21.54) 1019 0 R (lstnumber.-21.55) 1020 0 R (lstnumber.-21.56) 1021 0 R (lstnumber.-21.57) 1022 0 R] /Limits [(lstnumber.-21.52) (lstnumber.-21.57)] >> endobj 1457 0 obj << /Names [(lstnumber.-21.58) 1023 0 R (lstnumber.-21.59) 1024 0 R (lstnumber.-21.6) 967 0 R (lstnumber.-21.60) 1025 0 R (lstnumber.-21.61) 1026 0 R (lstnumber.-21.62) 1027 0 R] /Limits [(lstnumber.-21.58) (lstnumber.-21.62)] >> endobj 1458 0 obj << /Names [(lstnumber.-21.63) 1028 0 R (lstnumber.-21.64) 1029 0 R (lstnumber.-21.65) 1030 0 R (lstnumber.-21.66) 1031 0 R (lstnumber.-21.67) 1032 0 R (lstnumber.-21.68) 1033 0 R] /Limits [(lstnumber.-21.63) (lstnumber.-21.68)] >> endobj 1459 0 obj << /Names [(lstnumber.-21.69) 1034 0 R (lstnumber.-21.7) 968 0 R (lstnumber.-21.70) 1035 0 R (lstnumber.-21.71) 1036 0 R (lstnumber.-21.72) 1037 0 R (lstnumber.-21.73) 1038 0 R] /Limits [(lstnumber.-21.69) (lstnumber.-21.73)] >> endobj 1460 0 obj << /Names [(lstnumber.-21.74) 1039 0 R (lstnumber.-21.75) 1040 0 R (lstnumber.-21.76) 1041 0 R (lstnumber.-21.77) 1042 0 R (lstnumber.-21.78) 1043 0 R (lstnumber.-21.79) 1044 0 R] /Limits [(lstnumber.-21.74) (lstnumber.-21.79)] >> endobj 1461 0 obj << /Names [(lstnumber.-21.8) 969 0 R (lstnumber.-21.80) 1045 0 R (lstnumber.-21.81) 1046 0 R (lstnumber.-21.82) 1047 0 R (lstnumber.-21.83) 1048 0 R (lstnumber.-21.84) 1049 0 R] /Limits [(lstnumber.-21.8) (lstnumber.-21.84)] >> endobj 1462 0 obj << /Names [(lstnumber.-21.85) 1050 0 R (lstnumber.-21.86) 1051 0 R (lstnumber.-21.87) 1052 0 R (lstnumber.-21.88) 1053 0 R (lstnumber.-21.89) 1054 0 R (lstnumber.-21.9) 970 0 R] /Limits [(lstnumber.-21.85) (lstnumber.-21.9)] >> endobj 1463 0 obj << /Names [(lstnumber.-21.90) 1055 0 R (lstnumber.-21.91) 1056 0 R (lstnumber.-21.92) 1057 0 R (lstnumber.-21.93) 1058 0 R (lstnumber.-21.94) 1059 0 R (lstnumber.-21.95) 1060 0 R] /Limits [(lstnumber.-21.90) (lstnumber.-21.95)] >> endobj 1464 0 obj << /Names [(lstnumber.-21.96) 1061 0 R (lstnumber.-21.97) 1062 0 R (lstnumber.-21.98) 1063 0 R (lstnumber.-21.99) 1064 0 R (lstnumber.-22.1) 1091 0 R (lstnumber.-22.10) 1100 0 R] /Limits [(lstnumber.-21.96) (lstnumber.-22.10)] >> endobj 1465 0 obj << /Names [(lstnumber.-22.11) 1101 0 R (lstnumber.-22.12) 1102 0 R (lstnumber.-22.13) 1103 0 R (lstnumber.-22.14) 1104 0 R (lstnumber.-22.15) 1105 0 R (lstnumber.-22.16) 1106 0 R] /Limits [(lstnumber.-22.11) (lstnumber.-22.16)] >> endobj 1466 0 obj << /Names [(lstnumber.-22.17) 1107 0 R (lstnumber.-22.18) 1108 0 R (lstnumber.-22.19) 1109 0 R (lstnumber.-22.2) 1092 0 R (lstnumber.-22.20) 1110 0 R (lstnumber.-22.21) 1111 0 R] /Limits [(lstnumber.-22.17) (lstnumber.-22.21)] >> endobj 1467 0 obj << /Names [(lstnumber.-22.22) 1112 0 R (lstnumber.-22.23) 1113 0 R (lstnumber.-22.24) 1114 0 R (lstnumber.-22.25) 1115 0 R (lstnumber.-22.26) 1116 0 R (lstnumber.-22.27) 1117 0 R] /Limits [(lstnumber.-22.22) (lstnumber.-22.27)] >> endobj 1468 0 obj << /Names [(lstnumber.-22.28) 1118 0 R (lstnumber.-22.29) 1119 0 R (lstnumber.-22.3) 1093 0 R (lstnumber.-22.30) 1120 0 R (lstnumber.-22.31) 1121 0 R (lstnumber.-22.32) 1122 0 R] /Limits [(lstnumber.-22.28) (lstnumber.-22.32)] >> endobj 1469 0 obj << /Names [(lstnumber.-22.33) 1123 0 R (lstnumber.-22.34) 1124 0 R (lstnumber.-22.35) 1125 0 R (lstnumber.-22.36) 1126 0 R (lstnumber.-22.37) 1127 0 R (lstnumber.-22.38) 1128 0 R] /Limits [(lstnumber.-22.33) (lstnumber.-22.38)] >> endobj 1470 0 obj << /Names [(lstnumber.-22.39) 1129 0 R (lstnumber.-22.4) 1094 0 R (lstnumber.-22.40) 1130 0 R (lstnumber.-22.41) 1131 0 R (lstnumber.-22.42) 1132 0 R (lstnumber.-22.43) 1133 0 R] /Limits [(lstnumber.-22.39) (lstnumber.-22.43)] >> endobj 1471 0 obj << /Names [(lstnumber.-22.44) 1134 0 R (lstnumber.-22.45) 1135 0 R (lstnumber.-22.46) 1136 0 R (lstnumber.-22.47) 1137 0 R (lstnumber.-22.48) 1138 0 R (lstnumber.-22.49) 1139 0 R] /Limits [(lstnumber.-22.44) (lstnumber.-22.49)] >> endobj 1472 0 obj << /Names [(lstnumber.-22.5) 1095 0 R (lstnumber.-22.6) 1096 0 R (lstnumber.-22.7) 1097 0 R (lstnumber.-22.8) 1098 0 R (lstnumber.-22.9) 1099 0 R (lstnumber.-3.1) 317 0 R] /Limits [(lstnumber.-22.5) (lstnumber.-3.1)] >> endobj 1473 0 obj << /Names [(lstnumber.-4.1) 319 0 R (lstnumber.-4.2) 320 0 R (lstnumber.-4.3) 321 0 R (lstnumber.-5.1) 324 0 R (lstnumber.-5.2) 325 0 R (lstnumber.-5.3) 326 0 R] /Limits [(lstnumber.-4.1) (lstnumber.-5.3)] >> endobj 1474 0 obj << /Names [(lstnumber.-5.4) 327 0 R (lstnumber.-5.5) 328 0 R (lstnumber.-5.6) 329 0 R (lstnumber.-6.1) 332 0 R (lstnumber.-6.10) 341 0 R (lstnumber.-6.11) 342 0 R] /Limits [(lstnumber.-5.4) (lstnumber.-6.11)] >> endobj 1475 0 obj << /Names [(lstnumber.-6.12) 343 0 R (lstnumber.-6.13) 344 0 R (lstnumber.-6.14) 345 0 R (lstnumber.-6.15) 346 0 R (lstnumber.-6.2) 333 0 R (lstnumber.-6.3) 334 0 R] /Limits [(lstnumber.-6.12) (lstnumber.-6.3)] >> endobj 1476 0 obj << /Names [(lstnumber.-6.4) 335 0 R (lstnumber.-6.5) 336 0 R (lstnumber.-6.6) 337 0 R (lstnumber.-6.7) 338 0 R (lstnumber.-6.8) 339 0 R (lstnumber.-6.9) 340 0 R] /Limits [(lstnumber.-6.4) (lstnumber.-6.9)] >> endobj 1477 0 obj << /Names [(lstnumber.-7.1) 349 0 R (lstnumber.-7.10) 376 0 R (lstnumber.-7.11) 377 0 R (lstnumber.-7.12) 378 0 R (lstnumber.-7.2) 350 0 R (lstnumber.-7.3) 351 0 R] /Limits [(lstnumber.-7.1) (lstnumber.-7.3)] >> endobj 1478 0 obj << /Names [(lstnumber.-7.4) 352 0 R (lstnumber.-7.5) 353 0 R (lstnumber.-7.6) 354 0 R (lstnumber.-7.7) 373 0 R (lstnumber.-7.8) 374 0 R (lstnumber.-7.9) 375 0 R] /Limits [(lstnumber.-7.4) (lstnumber.-7.9)] >> endobj 1479 0 obj << /Names [(lstnumber.-8.1) 380 0 R (lstnumber.-8.10) 389 0 R (lstnumber.-8.11) 390 0 R (lstnumber.-8.12) 391 0 R (lstnumber.-8.13) 392 0 R (lstnumber.-8.14) 393 0 R] /Limits [(lstnumber.-8.1) (lstnumber.-8.14)] >> endobj 1480 0 obj << /Names [(lstnumber.-8.15) 394 0 R (lstnumber.-8.16) 395 0 R (lstnumber.-8.17) 396 0 R (lstnumber.-8.18) 397 0 R (lstnumber.-8.19) 398 0 R (lstnumber.-8.2) 381 0 R] /Limits [(lstnumber.-8.15) (lstnumber.-8.2)] >> endobj 1481 0 obj << /Names [(lstnumber.-8.20) 399 0 R (lstnumber.-8.21) 400 0 R (lstnumber.-8.22) 401 0 R (lstnumber.-8.23) 402 0 R (lstnumber.-8.24) 403 0 R (lstnumber.-8.25) 404 0 R] /Limits [(lstnumber.-8.20) (lstnumber.-8.25)] >> endobj 1482 0 obj << /Names [(lstnumber.-8.26) 405 0 R (lstnumber.-8.3) 382 0 R (lstnumber.-8.4) 383 0 R (lstnumber.-8.5) 384 0 R (lstnumber.-8.6) 385 0 R (lstnumber.-8.7) 386 0 R] /Limits [(lstnumber.-8.26) (lstnumber.-8.7)] >> endobj 1483 0 obj << /Names [(lstnumber.-8.8) 387 0 R (lstnumber.-8.9) 388 0 R (lstnumber.-9.1) 407 0 R (lstnumber.-9.10) 416 0 R (lstnumber.-9.11) 417 0 R (lstnumber.-9.12) 418 0 R] /Limits [(lstnumber.-8.8) (lstnumber.-9.12)] >> endobj 1484 0 obj << /Names [(lstnumber.-9.13) 427 0 R (lstnumber.-9.14) 428 0 R (lstnumber.-9.15) 429 0 R (lstnumber.-9.16) 430 0 R (lstnumber.-9.17) 431 0 R (lstnumber.-9.18) 432 0 R] /Limits [(lstnumber.-9.13) (lstnumber.-9.18)] >> endobj 1485 0 obj << /Names [(lstnumber.-9.19) 433 0 R (lstnumber.-9.2) 408 0 R (lstnumber.-9.20) 434 0 R (lstnumber.-9.21) 435 0 R (lstnumber.-9.22) 436 0 R (lstnumber.-9.23) 437 0 R] /Limits [(lstnumber.-9.19) (lstnumber.-9.23)] >> endobj 1486 0 obj << /Names [(lstnumber.-9.24) 438 0 R (lstnumber.-9.25) 439 0 R (lstnumber.-9.3) 409 0 R (lstnumber.-9.4) 410 0 R (lstnumber.-9.5) 411 0 R (lstnumber.-9.6) 412 0 R] /Limits [(lstnumber.-9.24) (lstnumber.-9.6)] >> endobj 1487 0 obj << /Names [(lstnumber.-9.7) 413 0 R (lstnumber.-9.8) 414 0 R (lstnumber.-9.9) 415 0 R (noweb.NW2i2LRH-10PmPC-1) 784 0 R (noweb.NW2i2LRH-1KNHNk-1) 359 0 R (noweb.NW2i2LRH-1ep4xx-1) 440 0 R] /Limits [(lstnumber.-9.7) (noweb.NW2i2LRH-1ep4xx-1)] >> endobj 1488 0 obj << /Names [(noweb.NW2i2LRH-1fgq0k-1) 330 0 R (noweb.NW2i2LRH-1oKncb-1) 357 0 R (noweb.NW2i2LRH-1sNM4H-1) 783 0 R (noweb.NW2i2LRH-1uBDkH-1) 275 0 R (noweb.NW2i2LRH-27nDss-1) 347 0 R (noweb.NW2i2LRH-2o3xLS-1) 276 0 R] /Limits [(noweb.NW2i2LRH-1fgq0k-1) (noweb.NW2i2LRH-2o3xLS-1)] >> endobj 1489 0 obj << /Names [(noweb.NW2i2LRH-2yIl95-1) 355 0 R (noweb.NW2i2LRH-2zwz7b-1) 723 0 R (noweb.NW2i2LRH-3Gkbo6-1) 315 0 R (noweb.NW2i2LRH-3dLroe-1) 362 0 R (noweb.NW2i2LRH-4BH61R-1) 360 0 R (noweb.NW2i2LRH-4GHEiM-1) 762 0 R] /Limits [(noweb.NW2i2LRH-2yIl95-1) (noweb.NW2i2LRH-4GHEiM-1)] >> endobj 1490 0 obj << /Names [(noweb.NW2i2LRH-4Mxm99-1) 361 0 R (noweb.NW2i2LRH-4XnUPC-1) 222 0 R (noweb.NW2i2LRH-4ZZkMy-1) 356 0 R (noweb.NW2i2LRH-CpvDB-1) 322 0 R (noweb.NW2i2LRH-FUd2L-1) 358 0 R (noweb.NW2i2LRH-KwRQ2-1) 234 0 R] /Limits [(noweb.NW2i2LRH-4Mxm99-1) (noweb.NW2i2LRH-KwRQ2-1)] >> endobj 1491 0 obj << /Names [(noweb.NW2i2LRH-OTtxn-1) 277 0 R (page.1) 144 0 R (page.10) 563 0 R (page.11) 573 0 R (page.12) 579 0 R (page.13) 589 0 R] /Limits [(noweb.NW2i2LRH-OTtxn-1) (page.13)] >> endobj 1492 0 obj << /Names [(page.14) 647 0 R (page.15) 694 0 R (page.16) 737 0 R (page.17) 794 0 R (page.18) 833 0 R (page.19) 837 0 R] /Limits [(page.14) (page.19)] >> endobj 1493 0 obj << /Names [(page.2) 188 0 R (page.20) 842 0 R (page.21) 846 0 R (page.22) 850 0 R (page.23) 856 0 R (page.24) 909 0 R] /Limits [(page.2) (page.24)] >> endobj 1494 0 obj << /Names [(page.25) 931 0 R (page.26) 960 0 R (page.27) 1008 0 R (page.28) 1071 0 R (page.29) 1089 0 R (page.3) 200 0 R] /Limits [(page.25) (page.3)] >> endobj 1495 0 obj << /Names [(page.4) 220 0 R (page.5) 308 0 R (page.6) 372 0 R (page.7) 426 0 R (page.8) 466 0 R (page.9) 515 0 R] /Limits [(page.4) (page.9)] >> endobj 1496 0 obj << /Names [(section*.1) 189 0 R (section.1) 6 0 R (section.2) 10 0 R (section.3) 14 0 R (section.4) 34 0 R (section.5) 106 0 R] /Limits [(section*.1) (section.5)] >> endobj 1497 0 obj << /Names [(section.6) 122 0 R (subsection.3.1) 18 0 R (subsection.3.2) 22 0 R (subsection.3.3) 26 0 R (subsection.3.4) 30 0 R (subsection.4.1) 38 0 R] /Limits [(section.6) (subsection.4.1)] >> endobj 1498 0 obj << /Names [(subsection.4.2) 42 0 R (subsection.4.3) 46 0 R (subsection.4.4) 50 0 R (subsection.4.5) 54 0 R (subsection.4.6) 74 0 R (subsection.4.7) 90 0 R] /Limits [(subsection.4.2) (subsection.4.7)] >> endobj 1499 0 obj << /Names [(subsection.4.8) 98 0 R (subsection.5.1) 110 0 R (subsection.5.2) 114 0 R (subsection.5.3) 118 0 R (subsection.6.1) 126 0 R (subsection.6.2) 130 0 R] /Limits [(subsection.4.8) (subsection.6.2)] >> endobj 1500 0 obj << /Names [(subsection.6.3) 134 0 R (subsubsection.4.5.1) 58 0 R (subsubsection.4.5.2) 62 0 R (subsubsection.4.5.3) 66 0 R (subsubsection.4.5.4) 70 0 R (subsubsection.4.6.1) 78 0 R] /Limits [(subsection.6.3) (subsubsection.4.6.1)] >> endobj 1501 0 obj << /Names [(subsubsection.4.6.2) 82 0 R (subsubsection.4.6.3) 86 0 R (subsubsection.4.7.1) 94 0 R (subsubsection.4.8.1) 102 0 R] /Limits [(subsubsection.4.6.2) (subsubsection.4.8.1)] >> endobj 1502 0 obj << /Kids [1371 0 R 1372 0 R 1373 0 R 1374 0 R 1375 0 R 1376 0 R] /Limits [(Doc-Start) (lstnumber.-10.1)] >> endobj 1503 0 obj << /Kids [1377 0 R 1378 0 R 1379 0 R 1380 0 R 1381 0 R 1382 0 R] /Limits [(lstnumber.-10.10) (lstnumber.-11.28)] >> endobj 1504 0 obj << /Kids [1383 0 R 1384 0 R 1385 0 R 1386 0 R 1387 0 R 1388 0 R] /Limits [(lstnumber.-11.29) (lstnumber.-12.23)] >> endobj 1505 0 obj << /Kids [1389 0 R 1390 0 R 1391 0 R 1392 0 R 1393 0 R 1394 0 R] /Limits [(lstnumber.-12.24) (lstnumber.-13.19)] >> endobj 1506 0 obj << /Kids [1395 0 R 1396 0 R 1397 0 R 1398 0 R 1399 0 R 1400 0 R] /Limits [(lstnumber.-13.2) (lstnumber.-13.51)] >> endobj 1507 0 obj << /Kids [1401 0 R 1402 0 R 1403 0 R 1404 0 R 1405 0 R 1406 0 R] /Limits [(lstnumber.-13.52) (lstnumber.-13.84)] >> endobj 1508 0 obj << /Kids [1407 0 R 1408 0 R 1409 0 R 1410 0 R 1411 0 R 1412 0 R] /Limits [(lstnumber.-13.85) (lstnumber.-14.4)] >> endobj 1509 0 obj << /Kids [1413 0 R 1414 0 R 1415 0 R 1416 0 R 1417 0 R 1418 0 R] /Limits [(lstnumber.-14.5) (lstnumber.-16.6)] >> endobj 1510 0 obj << /Kids [1419 0 R 1420 0 R 1421 0 R 1422 0 R 1423 0 R 1424 0 R] /Limits [(lstnumber.-16.7) (lstnumber.-18.9)] >> endobj 1511 0 obj << /Kids [1425 0 R 1426 0 R 1427 0 R 1428 0 R 1429 0 R 1430 0 R] /Limits [(lstnumber.-19.1) (lstnumber.-19.41)] >> endobj 1512 0 obj << /Kids [1431 0 R 1432 0 R 1433 0 R 1434 0 R 1435 0 R 1436 0 R] /Limits [(lstnumber.-19.42) (lstnumber.-2.31)] >> endobj 1513 0 obj << /Kids [1437 0 R 1438 0 R 1439 0 R 1440 0 R 1441 0 R 1442 0 R] /Limits [(lstnumber.-2.32) (lstnumber.-20.23)] >> endobj 1514 0 obj << /Kids [1443 0 R 1444 0 R 1445 0 R 1446 0 R 1447 0 R 1448 0 R] /Limits [(lstnumber.-20.24) (lstnumber.-21.13)] >> endobj 1515 0 obj << /Kids [1449 0 R 1450 0 R 1451 0 R 1452 0 R 1453 0 R 1454 0 R] /Limits [(lstnumber.-21.14) (lstnumber.-21.46)] >> endobj 1516 0 obj << /Kids [1455 0 R 1456 0 R 1457 0 R 1458 0 R 1459 0 R 1460 0 R] /Limits [(lstnumber.-21.47) (lstnumber.-21.79)] >> endobj 1517 0 obj << /Kids [1461 0 R 1462 0 R 1463 0 R 1464 0 R 1465 0 R 1466 0 R] /Limits [(lstnumber.-21.8) (lstnumber.-22.21)] >> endobj 1518 0 obj << /Kids [1467 0 R 1468 0 R 1469 0 R 1470 0 R 1471 0 R 1472 0 R] /Limits [(lstnumber.-22.22) (lstnumber.-3.1)] >> endobj 1519 0 obj << /Kids [1473 0 R 1474 0 R 1475 0 R 1476 0 R 1477 0 R 1478 0 R] /Limits [(lstnumber.-4.1) (lstnumber.-7.9)] >> endobj 1520 0 obj << /Kids [1479 0 R 1480 0 R 1481 0 R 1482 0 R 1483 0 R 1484 0 R] /Limits [(lstnumber.-8.1) (lstnumber.-9.18)] >> endobj 1521 0 obj << /Kids [1485 0 R 1486 0 R 1487 0 R 1488 0 R 1489 0 R 1490 0 R] /Limits [(lstnumber.-9.19) (noweb.NW2i2LRH-KwRQ2-1)] >> endobj 1522 0 obj << /Kids [1491 0 R 1492 0 R 1493 0 R 1494 0 R 1495 0 R 1496 0 R] /Limits [(noweb.NW2i2LRH-OTtxn-1) (section.5)] >> endobj 1523 0 obj << /Kids [1497 0 R 1498 0 R 1499 0 R 1500 0 R 1501 0 R] /Limits [(section.6) (subsubsection.4.8.1)] >> endobj 1524 0 obj << /Kids [1502 0 R 1503 0 R 1504 0 R 1505 0 R 1506 0 R 1507 0 R] /Limits [(Doc-Start) (lstnumber.-13.84)] >> endobj 1525 0 obj << /Kids [1508 0 R 1509 0 R 1510 0 R 1511 0 R 1512 0 R 1513 0 R] /Limits [(lstnumber.-13.85) (lstnumber.-20.23)] >> endobj 1526 0 obj << /Kids [1514 0 R 1515 0 R 1516 0 R 1517 0 R 1518 0 R 1519 0 R] /Limits [(lstnumber.-20.24) (lstnumber.-7.9)] >> endobj 1527 0 obj << /Kids [1520 0 R 1521 0 R 1522 0 R 1523 0 R] /Limits [(lstnumber.-8.1) (subsubsection.4.8.1)] >> endobj 1528 0 obj << /Kids [1524 0 R 1525 0 R 1526 0 R 1527 0 R] /Limits [(Doc-Start) (subsubsection.4.8.1)] >> endobj 1529 0 obj << /Dests 1528 0 R >> endobj 1530 0 obj << /Type /Catalog /Pages 1369 0 R /Outlines 1370 0 R /Names 1529 0 R /PageMode/UseOutlines /OpenAction 137 0 R >> endobj 1531 0 obj << /Author()/Title()/Subject()/Creator(LaTeX with hyperref package)/Producer(pdfTeX-1.40.9)/Keywords() /CreationDate (D:20110218160846Z) /ModDate (D:20110218160846Z) /Trapped /False /PTEX.Fullbanner (This is pdfTeX, Version 3.1415926-1.40.9-2.2 (Web2C 7.5.7) kpathsea version 3.5.7) >> endobj xref 0 1532 0000000001 65535 f 0000000002 00000 f 0000000003 00000 f 0000000004 00000 f 0000000000 00000 f 0000000015 00000 n 0000016720 00000 n 0000391260 00000 n 0000000060 00000 n 0000000118 00000 n 0000022641 00000 n 0000391173 00000 n 0000000163 00000 n 0000000197 00000 n 0000057062 00000 n 0000391047 00000 n 0000000243 00000 n 0000000279 00000 n 0000064678 00000 n 0000390973 00000 n 0000000330 00000 n 0000000365 00000 n 0000071884 00000 n 0000390886 00000 n 0000000416 00000 n 0000000458 00000 n 0000072001 00000 n 0000390799 00000 n 0000000509 00000 n 0000000551 00000 n 0000075150 00000 n 0000390725 00000 n 0000000602 00000 n 0000000638 00000 n 0000078980 00000 n 0000390598 00000 n 0000000684 00000 n 0000000723 00000 n 0000079036 00000 n 0000390524 00000 n 0000000774 00000 n 0000000803 00000 n 0000084097 00000 n 0000390437 00000 n 0000000854 00000 n 0000000882 00000 n 0000095180 00000 n 0000390350 00000 n 0000000933 00000 n 0000000984 00000 n 0000100029 00000 n 0000390263 00000 n 0000001035 00000 n 0000001109 00000 n 0000100085 00000 n 0000390139 00000 n 0000001160 00000 n 0000001198 00000 n 0000108420 00000 n 0000390065 00000 n 0000001254 00000 n 0000001288 00000 n 0000108475 00000 n 0000389978 00000 n 0000001344 00000 n 0000001386 00000 n 0000113917 00000 n 0000389891 00000 n 0000001442 00000 n 0000001484 00000 n 0000114945 00000 n 0000389817 00000 n 0000001540 00000 n 0000001576 00000 n 0000119814 00000 n 0000389693 00000 n 0000001627 00000 n 0000001664 00000 n 0000119870 00000 n 0000389619 00000 n 0000001720 00000 n 0000001762 00000 n 0000119926 00000 n 0000389532 00000 n 0000001818 00000 n 0000001860 00000 n 0000129353 00000 n 0000389458 00000 n 0000001916 00000 n 0000001952 00000 n 0000133343 00000 n 0000389334 00000 n 0000002003 00000 n 0000002044 00000 n 0000133399 00000 n 0000389273 00000 n 0000002100 00000 n 0000002142 00000 n 0000133455 00000 n 0000389159 00000 n 0000002193 00000 n 0000002243 00000 n 0000135053 00000 n 0000389095 00000 n 0000002300 00000 n 0000002343 00000 n 0000139225 00000 n 0000388963 00000 n 0000002390 00000 n 0000002425 00000 n 0000139282 00000 n 0000388884 00000 n 0000002477 00000 n 0000002506 00000 n 0000147214 00000 n 0000388791 00000 n 0000002558 00000 n 0000002604 00000 n 0000147271 00000 n 0000388712 00000 n 0000002656 00000 n 0000002711 00000 n 0000156802 00000 n 0000388593 00000 n 0000002758 00000 n 0000002798 00000 n 0000156859 00000 n 0000388514 00000 n 0000002850 00000 n 0000002879 00000 n 0000174889 00000 n 0000388421 00000 n 0000002931 00000 n 0000002983 00000 n 0000174947 00000 n 0000388342 00000 n 0000003035 00000 n 0000003096 00000 n 0000004604 00000 n 0000004752 00000 n 0000004932 00000 n 0000005142 00000 n 0000005441 00000 n 0000003148 00000 n 0000005327 00000 n 0000005384 00000 n 0000387247 00000 n 0000237148 00000 n 0000387078 00000 n 0000232927 00000 n 0000387416 00000 n 0000387586 00000 n 0000006935 00000 n 0000007086 00000 n 0000007236 00000 n 0000007387 00000 n 0000007543 00000 n 0000007699 00000 n 0000007855 00000 n 0000008010 00000 n 0000008160 00000 n 0000008315 00000 n 0000008471 00000 n 0000008627 00000 n 0000008783 00000 n 0000008939 00000 n 0000009101 00000 n 0000009263 00000 n 0000009424 00000 n 0000009586 00000 n 0000009741 00000 n 0000009901 00000 n 0000010062 00000 n 0000010224 00000 n 0000010379 00000 n 0000010541 00000 n 0000010697 00000 n 0000010859 00000 n 0000011010 00000 n 0000011166 00000 n 0000011322 00000 n 0000011478 00000 n 0000011627 00000 n 0000011783 00000 n 0000011939 00000 n 0000012214 00000 n 0000006547 00000 n 0000005565 00000 n 0000012095 00000 n 0000012157 00000 n 0000207813 00000 n 0000015751 00000 n 0000015903 00000 n 0000016055 00000 n 0000016207 00000 n 0000016359 00000 n 0000016511 00000 n 0000016836 00000 n 0000015579 00000 n 0000012312 00000 n 0000016663 00000 n 0000386208 00000 n 0000386353 00000 n 0000385632 00000 n 0000386497 00000 n 0000016775 00000 n 0000385052 00000 n 0000385777 00000 n 0000021094 00000 n 0000021258 00000 n 0000021423 00000 n 0000021588 00000 n 0000021754 00000 n 0000021920 00000 n 0000022085 00000 n 0000022248 00000 n 0000022413 00000 n 0000025851 00000 n 0000020898 00000 n 0000017012 00000 n 0000022579 00000 n 0000188044 00000 n 0000022697 00000 n 0000385197 00000 n 0000022757 00000 n 0000022814 00000 n 0000022874 00000 n 0000022935 00000 n 0000022994 00000 n 0000023055 00000 n 0000023116 00000 n 0000023177 00000 n 0000023238 00000 n 0000023299 00000 n 0000023360 00000 n 0000023421 00000 n 0000023478 00000 n 0000023539 00000 n 0000023600 00000 n 0000023661 00000 n 0000023722 00000 n 0000023783 00000 n 0000023844 00000 n 0000023905 00000 n 0000023966 00000 n 0000024027 00000 n 0000024088 00000 n 0000024149 00000 n 0000024210 00000 n 0000024271 00000 n 0000024332 00000 n 0000024393 00000 n 0000024454 00000 n 0000024515 00000 n 0000024576 00000 n 0000024637 00000 n 0000024698 00000 n 0000024759 00000 n 0000024820 00000 n 0000024881 00000 n 0000024942 00000 n 0000025003 00000 n 0000025063 00000 n 0000025124 00000 n 0000025184 00000 n 0000025245 00000 n 0000025305 00000 n 0000025366 00000 n 0000025427 00000 n 0000025488 00000 n 0000025549 00000 n 0000025610 00000 n 0000025671 00000 n 0000025731 00000 n 0000025791 00000 n 0000035732 00000 n 0000044075 00000 n 0000045775 00000 n 0000030676 00000 n 0000030842 00000 n 0000031006 00000 n 0000031172 00000 n 0000031336 00000 n 0000031501 00000 n 0000031667 00000 n 0000031833 00000 n 0000031998 00000 n 0000032164 00000 n 0000032330 00000 n 0000032496 00000 n 0000032662 00000 n 0000032826 00000 n 0000032992 00000 n 0000033158 00000 n 0000033323 00000 n 0000033489 00000 n 0000033654 00000 n 0000033817 00000 n 0000033981 00000 n 0000034146 00000 n 0000034310 00000 n 0000034475 00000 n 0000034641 00000 n 0000034805 00000 n 0000034970 00000 n 0000038026 00000 n 0000030336 00000 n 0000026001 00000 n 0000035136 00000 n 0000035193 00000 n 0000035254 00000 n 0000035315 00000 n 0000035375 00000 n 0000035436 00000 n 0000035496 00000 n 0000035557 00000 n 0000035616 00000 n 0000035673 00000 n 0000035793 00000 n 0000035850 00000 n 0000035911 00000 n 0000035972 00000 n 0000036032 00000 n 0000036093 00000 n 0000036150 00000 n 0000036211 00000 n 0000036272 00000 n 0000036333 00000 n 0000036394 00000 n 0000036455 00000 n 0000036516 00000 n 0000036577 00000 n 0000036634 00000 n 0000036695 00000 n 0000036756 00000 n 0000036817 00000 n 0000036878 00000 n 0000036938 00000 n 0000036999 00000 n 0000037059 00000 n 0000037120 00000 n 0000037180 00000 n 0000037241 00000 n 0000037302 00000 n 0000037363 00000 n 0000037424 00000 n 0000037485 00000 n 0000037546 00000 n 0000037607 00000 n 0000037663 00000 n 0000037724 00000 n 0000037785 00000 n 0000037846 00000 n 0000037906 00000 n 0000037966 00000 n 0000057118 00000 n 0000064734 00000 n 0000084153 00000 n 0000100141 00000 n 0000139339 00000 n 0000147327 00000 n 0000156915 00000 n 0000180083 00000 n 0000042663 00000 n 0000042826 00000 n 0000042992 00000 n 0000043158 00000 n 0000043320 00000 n 0000043484 00000 n 0000046622 00000 n 0000042491 00000 n 0000038150 00000 n 0000043649 00000 n 0000043711 00000 n 0000043772 00000 n 0000043833 00000 n 0000043893 00000 n 0000043954 00000 n 0000044014 00000 n 0000044136 00000 n 0000044193 00000 n 0000044254 00000 n 0000044315 00000 n 0000044376 00000 n 0000044436 00000 n 0000044497 00000 n 0000044556 00000 n 0000044617 00000 n 0000044677 00000 n 0000044738 00000 n 0000044799 00000 n 0000044860 00000 n 0000044921 00000 n 0000044982 00000 n 0000045043 00000 n 0000045104 00000 n 0000045165 00000 n 0000045226 00000 n 0000045287 00000 n 0000045348 00000 n 0000045409 00000 n 0000045470 00000 n 0000045531 00000 n 0000045592 00000 n 0000045653 00000 n 0000045714 00000 n 0000045836 00000 n 0000045893 00000 n 0000045954 00000 n 0000046015 00000 n 0000046076 00000 n 0000046137 00000 n 0000046198 00000 n 0000046259 00000 n 0000046320 00000 n 0000046381 00000 n 0000046442 00000 n 0000046502 00000 n 0000046562 00000 n 0000049705 00000 n 0000049871 00000 n 0000050035 00000 n 0000050199 00000 n 0000052303 00000 n 0000049549 00000 n 0000046746 00000 n 0000050362 00000 n 0000050419 00000 n 0000050480 00000 n 0000050541 00000 n 0000050601 00000 n 0000050662 00000 n 0000050722 00000 n 0000050783 00000 n 0000050843 00000 n 0000050904 00000 n 0000050965 00000 n 0000051026 00000 n 0000051087 00000 n 0000051148 00000 n 0000051209 00000 n 0000051270 00000 n 0000051327 00000 n 0000051388 00000 n 0000051449 00000 n 0000051510 00000 n 0000051571 00000 n 0000051632 00000 n 0000051693 00000 n 0000051754 00000 n 0000051815 00000 n 0000051876 00000 n 0000051937 00000 n 0000051998 00000 n 0000052059 00000 n 0000052120 00000 n 0000052181 00000 n 0000052242 00000 n 0000387704 00000 n 0000056505 00000 n 0000056669 00000 n 0000056834 00000 n 0000063974 00000 n 0000059731 00000 n 0000056357 00000 n 0000052414 00000 n 0000057000 00000 n 0000057179 00000 n 0000057236 00000 n 0000057297 00000 n 0000057358 00000 n 0000057419 00000 n 0000057480 00000 n 0000057541 00000 n 0000057602 00000 n 0000057663 00000 n 0000057724 00000 n 0000057785 00000 n 0000057846 00000 n 0000057907 00000 n 0000057968 00000 n 0000058029 00000 n 0000058090 00000 n 0000058151 00000 n 0000058212 00000 n 0000058273 00000 n 0000058334 00000 n 0000058394 00000 n 0000058455 00000 n 0000058515 00000 n 0000058576 00000 n 0000058633 00000 n 0000058694 00000 n 0000058755 00000 n 0000058816 00000 n 0000058877 00000 n 0000058938 00000 n 0000058999 00000 n 0000059060 00000 n 0000059121 00000 n 0000059182 00000 n 0000059243 00000 n 0000059304 00000 n 0000059365 00000 n 0000059426 00000 n 0000059487 00000 n 0000059548 00000 n 0000059609 00000 n 0000059670 00000 n 0000064126 00000 n 0000064290 00000 n 0000064455 00000 n 0000067350 00000 n 0000063818 00000 n 0000059868 00000 n 0000064621 00000 n 0000064795 00000 n 0000064852 00000 n 0000064913 00000 n 0000064974 00000 n 0000065035 00000 n 0000065096 00000 n 0000065157 00000 n 0000065218 00000 n 0000065279 00000 n 0000065340 00000 n 0000065401 00000 n 0000065462 00000 n 0000065523 00000 n 0000065584 00000 n 0000065645 00000 n 0000065706 00000 n 0000065767 00000 n 0000065828 00000 n 0000065889 00000 n 0000065950 00000 n 0000066011 00000 n 0000066072 00000 n 0000066132 00000 n 0000066193 00000 n 0000066253 00000 n 0000066314 00000 n 0000066374 00000 n 0000066435 00000 n 0000066496 00000 n 0000066557 00000 n 0000066618 00000 n 0000066679 00000 n 0000066740 00000 n 0000066801 00000 n 0000066862 00000 n 0000066923 00000 n 0000066984 00000 n 0000067045 00000 n 0000067106 00000 n 0000067167 00000 n 0000067228 00000 n 0000067289 00000 n 0000071521 00000 n 0000071672 00000 n 0000072057 00000 n 0000071381 00000 n 0000067539 00000 n 0000071822 00000 n 0000071940 00000 n 0000385342 00000 n 0000385486 00000 n 0000074345 00000 n 0000074551 00000 n 0000074941 00000 n 0000075206 00000 n 0000074189 00000 n 0000072272 00000 n 0000075093 00000 n 0000386788 00000 n 0000074746 00000 n 0000079154 00000 n 0000078806 00000 n 0000075421 00000 n 0000078918 00000 n 0000386933 00000 n 0000386065 00000 n 0000385921 00000 n 0000079092 00000 n 0000083710 00000 n 0000083874 00000 n 0000087373 00000 n 0000083570 00000 n 0000079397 00000 n 0000084040 00000 n 0000084214 00000 n 0000084271 00000 n 0000084332 00000 n 0000084393 00000 n 0000084454 00000 n 0000084515 00000 n 0000084576 00000 n 0000084637 00000 n 0000084698 00000 n 0000084759 00000 n 0000084820 00000 n 0000084881 00000 n 0000084942 00000 n 0000085003 00000 n 0000085064 00000 n 0000085125 00000 n 0000085186 00000 n 0000085247 00000 n 0000085308 00000 n 0000085369 00000 n 0000085430 00000 n 0000085491 00000 n 0000085552 00000 n 0000085613 00000 n 0000085674 00000 n 0000085731 00000 n 0000085792 00000 n 0000085852 00000 n 0000085913 00000 n 0000085973 00000 n 0000086034 00000 n 0000086095 00000 n 0000086156 00000 n 0000086217 00000 n 0000086278 00000 n 0000086339 00000 n 0000086400 00000 n 0000086461 00000 n 0000086522 00000 n 0000086583 00000 n 0000086644 00000 n 0000086705 00000 n 0000086766 00000 n 0000086827 00000 n 0000086888 00000 n 0000086949 00000 n 0000087010 00000 n 0000087071 00000 n 0000087132 00000 n 0000087193 00000 n 0000087253 00000 n 0000087313 00000 n 0000387822 00000 n 0000092640 00000 n 0000095236 00000 n 0000092508 00000 n 0000087497 00000 n 0000092806 00000 n 0000092868 00000 n 0000092929 00000 n 0000092990 00000 n 0000093050 00000 n 0000093111 00000 n 0000093171 00000 n 0000093232 00000 n 0000093292 00000 n 0000093353 00000 n 0000093414 00000 n 0000093475 00000 n 0000093536 00000 n 0000093597 00000 n 0000093658 00000 n 0000093719 00000 n 0000093780 00000 n 0000093841 00000 n 0000093902 00000 n 0000093963 00000 n 0000094024 00000 n 0000094085 00000 n 0000094146 00000 n 0000094207 00000 n 0000094268 00000 n 0000094329 00000 n 0000094390 00000 n 0000094451 00000 n 0000094512 00000 n 0000094573 00000 n 0000094634 00000 n 0000094695 00000 n 0000094756 00000 n 0000094816 00000 n 0000094877 00000 n 0000094937 00000 n 0000094998 00000 n 0000095058 00000 n 0000095119 00000 n 0000386643 00000 n 0000099314 00000 n 0000099477 00000 n 0000099642 00000 n 0000099808 00000 n 0000101901 00000 n 0000099158 00000 n 0000095531 00000 n 0000099972 00000 n 0000100202 00000 n 0000100259 00000 n 0000100320 00000 n 0000100381 00000 n 0000100442 00000 n 0000100503 00000 n 0000100564 00000 n 0000100625 00000 n 0000100686 00000 n 0000100747 00000 n 0000100808 00000 n 0000100869 00000 n 0000100930 00000 n 0000100991 00000 n 0000101052 00000 n 0000101113 00000 n 0000101174 00000 n 0000101235 00000 n 0000101296 00000 n 0000101357 00000 n 0000101418 00000 n 0000101479 00000 n 0000101540 00000 n 0000101601 00000 n 0000101662 00000 n 0000101722 00000 n 0000101783 00000 n 0000101841 00000 n 0000107390 00000 n 0000105201 00000 n 0000105367 00000 n 0000105533 00000 n 0000105699 00000 n 0000105862 00000 n 0000106028 00000 n 0000106193 00000 n 0000106350 00000 n 0000106512 00000 n 0000106678 00000 n 0000109800 00000 n 0000104997 00000 n 0000102038 00000 n 0000106843 00000 n 0000106905 00000 n 0000106966 00000 n 0000107027 00000 n 0000107087 00000 n 0000107148 00000 n 0000107208 00000 n 0000107269 00000 n 0000107329 00000 n 0000107451 00000 n 0000107508 00000 n 0000107569 00000 n 0000107630 00000 n 0000107691 00000 n 0000107752 00000 n 0000107813 00000 n 0000107874 00000 n 0000107935 00000 n 0000107996 00000 n 0000108057 00000 n 0000108118 00000 n 0000108178 00000 n 0000108239 00000 n 0000108299 00000 n 0000108360 00000 n 0000108529 00000 n 0000108590 00000 n 0000108647 00000 n 0000108708 00000 n 0000108769 00000 n 0000108830 00000 n 0000108891 00000 n 0000108952 00000 n 0000109013 00000 n 0000109074 00000 n 0000109133 00000 n 0000109194 00000 n 0000109254 00000 n 0000109315 00000 n 0000109375 00000 n 0000109436 00000 n 0000109497 00000 n 0000109558 00000 n 0000109619 00000 n 0000109680 00000 n 0000109740 00000 n 0000113973 00000 n 0000115001 00000 n 0000112872 00000 n 0000113035 00000 n 0000113200 00000 n 0000113365 00000 n 0000113529 00000 n 0000113695 00000 n 0000116273 00000 n 0000112700 00000 n 0000109950 00000 n 0000113860 00000 n 0000114034 00000 n 0000114091 00000 n 0000114152 00000 n 0000114213 00000 n 0000114274 00000 n 0000114335 00000 n 0000114396 00000 n 0000114457 00000 n 0000114518 00000 n 0000114579 00000 n 0000114640 00000 n 0000114701 00000 n 0000114762 00000 n 0000114823 00000 n 0000114884 00000 n 0000115061 00000 n 0000115118 00000 n 0000115178 00000 n 0000115239 00000 n 0000115299 00000 n 0000115360 00000 n 0000115419 00000 n 0000115480 00000 n 0000115541 00000 n 0000115602 00000 n 0000115663 00000 n 0000115724 00000 n 0000115785 00000 n 0000115846 00000 n 0000115907 00000 n 0000115968 00000 n 0000116029 00000 n 0000116090 00000 n 0000116151 00000 n 0000116212 00000 n 0000119982 00000 n 0000119640 00000 n 0000116397 00000 n 0000119752 00000 n 0000125386 00000 n 0000125217 00000 n 0000120197 00000 n 0000125329 00000 n 0000387940 00000 n 0000129409 00000 n 0000129179 00000 n 0000125588 00000 n 0000129291 00000 n 0000133511 00000 n 0000133174 00000 n 0000129598 00000 n 0000133286 00000 n 0000135110 00000 n 0000134879 00000 n 0000133700 00000 n 0000134991 00000 n 0000138840 00000 n 0000139004 00000 n 0000142136 00000 n 0000138700 00000 n 0000135286 00000 n 0000139168 00000 n 0000139400 00000 n 0000139457 00000 n 0000139518 00000 n 0000139579 00000 n 0000139640 00000 n 0000139701 00000 n 0000139762 00000 n 0000139823 00000 n 0000139884 00000 n 0000139945 00000 n 0000140006 00000 n 0000140067 00000 n 0000140128 00000 n 0000140189 00000 n 0000140250 00000 n 0000140311 00000 n 0000140372 00000 n 0000140433 00000 n 0000140494 00000 n 0000140555 00000 n 0000140615 00000 n 0000140676 00000 n 0000140736 00000 n 0000140797 00000 n 0000140857 00000 n 0000140918 00000 n 0000140979 00000 n 0000141040 00000 n 0000141101 00000 n 0000141162 00000 n 0000141223 00000 n 0000141284 00000 n 0000141345 00000 n 0000141406 00000 n 0000141467 00000 n 0000141528 00000 n 0000141589 00000 n 0000141650 00000 n 0000141711 00000 n 0000141772 00000 n 0000141833 00000 n 0000141894 00000 n 0000141955 00000 n 0000142016 00000 n 0000142076 00000 n 0000146309 00000 n 0000146475 00000 n 0000146639 00000 n 0000146805 00000 n 0000148297 00000 n 0000146153 00000 n 0000142366 00000 n 0000146970 00000 n 0000147032 00000 n 0000147093 00000 n 0000147154 00000 n 0000147388 00000 n 0000147445 00000 n 0000147506 00000 n 0000147567 00000 n 0000147628 00000 n 0000147689 00000 n 0000147750 00000 n 0000147811 00000 n 0000147872 00000 n 0000147933 00000 n 0000147994 00000 n 0000148055 00000 n 0000148116 00000 n 0000148177 00000 n 0000148237 00000 n 0000151683 00000 n 0000150175 00000 n 0000148525 00000 n 0000150287 00000 n 0000150344 00000 n 0000150405 00000 n 0000150466 00000 n 0000150526 00000 n 0000150587 00000 n 0000150647 00000 n 0000150708 00000 n 0000150768 00000 n 0000150829 00000 n 0000150890 00000 n 0000150951 00000 n 0000151012 00000 n 0000151073 00000 n 0000151134 00000 n 0000151195 00000 n 0000151256 00000 n 0000151317 00000 n 0000151378 00000 n 0000151439 00000 n 0000151500 00000 n 0000151561 00000 n 0000151622 00000 n 0000388058 00000 n 0000156410 00000 n 0000156574 00000 n 0000159655 00000 n 0000156270 00000 n 0000151768 00000 n 0000156740 00000 n 0000156976 00000 n 0000157033 00000 n 0000157094 00000 n 0000157155 00000 n 0000157216 00000 n 0000157277 00000 n 0000157338 00000 n 0000157399 00000 n 0000157460 00000 n 0000157521 00000 n 0000157582 00000 n 0000157643 00000 n 0000157704 00000 n 0000157765 00000 n 0000157826 00000 n 0000157887 00000 n 0000157948 00000 n 0000158009 00000 n 0000158070 00000 n 0000158131 00000 n 0000158192 00000 n 0000158253 00000 n 0000158314 00000 n 0000158375 00000 n 0000158436 00000 n 0000158497 00000 n 0000158557 00000 n 0000158618 00000 n 0000158678 00000 n 0000158739 00000 n 0000158799 00000 n 0000158860 00000 n 0000158921 00000 n 0000158982 00000 n 0000159043 00000 n 0000159104 00000 n 0000159165 00000 n 0000159226 00000 n 0000159287 00000 n 0000159348 00000 n 0000159410 00000 n 0000159472 00000 n 0000159533 00000 n 0000159594 00000 n 0000167145 00000 n 0000163326 00000 n 0000159885 00000 n 0000163441 00000 n 0000163500 00000 n 0000163563 00000 n 0000163626 00000 n 0000163688 00000 n 0000163751 00000 n 0000163813 00000 n 0000163876 00000 n 0000163938 00000 n 0000164001 00000 n 0000164064 00000 n 0000164127 00000 n 0000164190 00000 n 0000164253 00000 n 0000164316 00000 n 0000164379 00000 n 0000164442 00000 n 0000164505 00000 n 0000164568 00000 n 0000164631 00000 n 0000164694 00000 n 0000164757 00000 n 0000164820 00000 n 0000164883 00000 n 0000164946 00000 n 0000165009 00000 n 0000165072 00000 n 0000165135 00000 n 0000165198 00000 n 0000165261 00000 n 0000165324 00000 n 0000165387 00000 n 0000165450 00000 n 0000165512 00000 n 0000165575 00000 n 0000165637 00000 n 0000165700 00000 n 0000165762 00000 n 0000165825 00000 n 0000165888 00000 n 0000165951 00000 n 0000166014 00000 n 0000166077 00000 n 0000166140 00000 n 0000166203 00000 n 0000166266 00000 n 0000166329 00000 n 0000166392 00000 n 0000166455 00000 n 0000166518 00000 n 0000166581 00000 n 0000166644 00000 n 0000166707 00000 n 0000166770 00000 n 0000166833 00000 n 0000166896 00000 n 0000166959 00000 n 0000167021 00000 n 0000167083 00000 n 0000174094 00000 n 0000175005 00000 n 0000173958 00000 n 0000167231 00000 n 0000174261 00000 n 0000174325 00000 n 0000174388 00000 n 0000174451 00000 n 0000174513 00000 n 0000174576 00000 n 0000174638 00000 n 0000174701 00000 n 0000174763 00000 n 0000174826 00000 n 0000184293 00000 n 0000183639 00000 n 0000179525 00000 n 0000179690 00000 n 0000179857 00000 n 0000183285 00000 n 0000179371 00000 n 0000175331 00000 n 0000180024 00000 n 0000180145 00000 n 0000180204 00000 n 0000180267 00000 n 0000180330 00000 n 0000180393 00000 n 0000180456 00000 n 0000180518 00000 n 0000180581 00000 n 0000180643 00000 n 0000180706 00000 n 0000180768 00000 n 0000180831 00000 n 0000180894 00000 n 0000180957 00000 n 0000181020 00000 n 0000181083 00000 n 0000181146 00000 n 0000181209 00000 n 0000181272 00000 n 0000181335 00000 n 0000181398 00000 n 0000181461 00000 n 0000181524 00000 n 0000181587 00000 n 0000181650 00000 n 0000181713 00000 n 0000181776 00000 n 0000181839 00000 n 0000181902 00000 n 0000181965 00000 n 0000182028 00000 n 0000182091 00000 n 0000182154 00000 n 0000182217 00000 n 0000182280 00000 n 0000182343 00000 n 0000182406 00000 n 0000182468 00000 n 0000182531 00000 n 0000182593 00000 n 0000182656 00000 n 0000182718 00000 n 0000182781 00000 n 0000182844 00000 n 0000182907 00000 n 0000182970 00000 n 0000183033 00000 n 0000183096 00000 n 0000183159 00000 n 0000183222 00000 n 0000183410 00000 n 0000183894 00000 n 0000183921 00000 n 0000183985 00000 n 0000184024 00000 n 0000184544 00000 n 0000184571 00000 n 0000184635 00000 n 0000184674 00000 n 0000184881 00000 n 0000185075 00000 n 0000185272 00000 n 0000185398 00000 n 0000185570 00000 n 0000185967 00000 n 0000186655 00000 n 0000187298 00000 n 0000187703 00000 n 0000188298 00000 n 0000188347 00000 n 0000188432 00000 n 0000188486 00000 n 0000188634 00000 n 0000189336 00000 n 0000189542 00000 n 0000189858 00000 n 0000190533 00000 n 0000191202 00000 n 0000191481 00000 n 0000191759 00000 n 0000192038 00000 n 0000192217 00000 n 0000192493 00000 n 0000192685 00000 n 0000192879 00000 n 0000193141 00000 n 0000193402 00000 n 0000193627 00000 n 0000193851 00000 n 0000194029 00000 n 0000194273 00000 n 0000194459 00000 n 0000194652 00000 n 0000194988 00000 n 0000195260 00000 n 0000195540 00000 n 0000195852 00000 n 0000196131 00000 n 0000196362 00000 n 0000196674 00000 n 0000196861 00000 n 0000197069 00000 n 0000197352 00000 n 0000197633 00000 n 0000197899 00000 n 0000198163 00000 n 0000198462 00000 n 0000198762 00000 n 0000199094 00000 n 0000199303 00000 n 0000199559 00000 n 0000199838 00000 n 0000200106 00000 n 0000200408 00000 n 0000200696 00000 n 0000200983 00000 n 0000201285 00000 n 0000201583 00000 n 0000201829 00000 n 0000202186 00000 n 0000202433 00000 n 0000202652 00000 n 0000202951 00000 n 0000203156 00000 n 0000203399 00000 n 0000203640 00000 n 0000203920 00000 n 0000204223 00000 n 0000204477 00000 n 0000204777 00000 n 0000205019 00000 n 0000205260 00000 n 0000205516 00000 n 0000205793 00000 n 0000206081 00000 n 0000206400 00000 n 0000206653 00000 n 0000206953 00000 n 0000207178 00000 n 0000207492 00000 n 0000208062 00000 n 0000208509 00000 n 0000209029 00000 n 0000384681 00000 n 0000209959 00000 n 0000210437 00000 n 0000210656 00000 n 0000210870 00000 n 0000211144 00000 n 0000211413 00000 n 0000211662 00000 n 0000211840 00000 n 0000212084 00000 n 0000212280 00000 n 0000212478 00000 n 0000212712 00000 n 0000212969 00000 n 0000213222 00000 n 0000213408 00000 n 0000213638 00000 n 0000213891 00000 n 0000214089 00000 n 0000214307 00000 n 0000214488 00000 n 0000214727 00000 n 0000214916 00000 n 0000215153 00000 n 0000215345 00000 n 0000215530 00000 n 0000215729 00000 n 0000216058 00000 n 0000216379 00000 n 0000216724 00000 n 0000216997 00000 n 0000217263 00000 n 0000217535 00000 n 0000217835 00000 n 0000218100 00000 n 0000218329 00000 n 0000218556 00000 n 0000218865 00000 n 0000219067 00000 n 0000219257 00000 n 0000219557 00000 n 0000219762 00000 n 0000220021 00000 n 0000220297 00000 n 0000220549 00000 n 0000220805 00000 n 0000221096 00000 n 0000221375 00000 n 0000221693 00000 n 0000221899 00000 n 0000222148 00000 n 0000222419 00000 n 0000222698 00000 n 0000222972 00000 n 0000223236 00000 n 0000223509 00000 n 0000223778 00000 n 0000224042 00000 n 0000224317 00000 n 0000224600 00000 n 0000224836 00000 n 0000225171 00000 n 0000225414 00000 n 0000225628 00000 n 0000225879 00000 n 0000226159 00000 n 0000226356 00000 n 0000226609 00000 n 0000226846 00000 n 0000227111 00000 n 0000227393 00000 n 0000227686 00000 n 0000227926 00000 n 0000228193 00000 n 0000228430 00000 n 0000228660 00000 n 0000228928 00000 n 0000229184 00000 n 0000229468 00000 n 0000229787 00000 n 0000230045 00000 n 0000230330 00000 n 0000230553 00000 n 0000230857 00000 n 0000231165 00000 n 0000231428 00000 n 0000231717 00000 n 0000232040 00000 n 0000232305 00000 n 0000232600 00000 n 0000233181 00000 n 0000233740 00000 n 0000234233 00000 n 0000235516 00000 n 0000235870 00000 n 0000236264 00000 n 0000236486 00000 n 0000236839 00000 n 0000237399 00000 n 0000237456 00000 n 0000237558 00000 n 0000237638 00000 n 0000238542 00000 n 0000243667 00000 n 0000244305 00000 n 0000247818 00000 n 0000248208 00000 n 0000250444 00000 n 0000250742 00000 n 0000254264 00000 n 0000254643 00000 n 0000257074 00000 n 0000257363 00000 n 0000259500 00000 n 0000259755 00000 n 0000262372 00000 n 0000262612 00000 n 0000265218 00000 n 0000265455 00000 n 0000277922 00000 n 0000278319 00000 n 0000288138 00000 n 0000288480 00000 n 0000292066 00000 n 0000292312 00000 n 0000294593 00000 n 0000294813 00000 n 0000296741 00000 n 0000296983 00000 n 0000299706 00000 n 0000299931 00000 n 0000327495 00000 n 0000327858 00000 n 0000357555 00000 n 0000358051 00000 n 0000384344 00000 n 0000388171 00000 n 0000388266 00000 n 0000391333 00000 n 0000391543 00000 n 0000391770 00000 n 0000391996 00000 n 0000392218 00000 n 0000392442 00000 n 0000392671 00000 n 0000392914 00000 n 0000393151 00000 n 0000393389 00000 n 0000393632 00000 n 0000393874 00000 n 0000394117 00000 n 0000394359 00000 n 0000394602 00000 n 0000394839 00000 n 0000395078 00000 n 0000395321 00000 n 0000395563 00000 n 0000395806 00000 n 0000396047 00000 n 0000396288 00000 n 0000396526 00000 n 0000396766 00000 n 0000397009 00000 n 0000397250 00000 n 0000397491 00000 n 0000397734 00000 n 0000397976 00000 n 0000398219 00000 n 0000398461 00000 n 0000398704 00000 n 0000398946 00000 n 0000399189 00000 n 0000399431 00000 n 0000399674 00000 n 0000399915 00000 n 0000400156 00000 n 0000400397 00000 n 0000400638 00000 n 0000400881 00000 n 0000401123 00000 n 0000401364 00000 n 0000401599 00000 n 0000401842 00000 n 0000402077 00000 n 0000402316 00000 n 0000402559 00000 n 0000402796 00000 n 0000403034 00000 n 0000403273 00000 n 0000403508 00000 n 0000403751 00000 n 0000403991 00000 n 0000404226 00000 n 0000404467 00000 n 0000404708 00000 n 0000404951 00000 n 0000405193 00000 n 0000405436 00000 n 0000405678 00000 n 0000405921 00000 n 0000406154 00000 n 0000406389 00000 n 0000406623 00000 n 0000406858 00000 n 0000407092 00000 n 0000407327 00000 n 0000407561 00000 n 0000407792 00000 n 0000408028 00000 n 0000408271 00000 n 0000408513 00000 n 0000408756 00000 n 0000408997 00000 n 0000409235 00000 n 0000409479 00000 n 0000409736 00000 n 0000409986 00000 n 0000410229 00000 n 0000410470 00000 n 0000410711 00000 n 0000410954 00000 n 0000411198 00000 n 0000411447 00000 n 0000411694 00000 n 0000411943 00000 n 0000412190 00000 n 0000412439 00000 n 0000412686 00000 n 0000412935 00000 n 0000413181 00000 n 0000413427 00000 n 0000413676 00000 n 0000413924 00000 n 0000414173 00000 n 0000414421 00000 n 0000414670 00000 n 0000414918 00000 n 0000415167 00000 n 0000415415 00000 n 0000415664 00000 n 0000415902 00000 n 0000416129 00000 n 0000416359 00000 n 0000416591 00000 n 0000416818 00000 n 0000417048 00000 n 0000417275 00000 n 0000417508 00000 n 0000417741 00000 n 0000417976 00000 n 0000418205 00000 n 0000418436 00000 n 0000418671 00000 n 0000418905 00000 n 0000419135 00000 n 0000419398 00000 n 0000419697 00000 n 0000419996 00000 n 0000420291 00000 n 0000420491 00000 n 0000420662 00000 n 0000420831 00000 n 0000421003 00000 n 0000421166 00000 n 0000421350 00000 n 0000421562 00000 n 0000421783 00000 n 0000422009 00000 n 0000422261 00000 n 0000422465 00000 n 0000422591 00000 n 0000422725 00000 n 0000422859 00000 n 0000422993 00000 n 0000423126 00000 n 0000423260 00000 n 0000423393 00000 n 0000423525 00000 n 0000423657 00000 n 0000423790 00000 n 0000423923 00000 n 0000424056 00000 n 0000424190 00000 n 0000424324 00000 n 0000424458 00000 n 0000424591 00000 n 0000424723 00000 n 0000424853 00000 n 0000424984 00000 n 0000425123 00000 n 0000425256 00000 n 0000425377 00000 n 0000425504 00000 n 0000425638 00000 n 0000425770 00000 n 0000425887 00000 n 0000425999 00000 n 0000426039 00000 n 0000426171 00000 n trailer << /Size 1532 /Root 1530 0 R /Info 1531 0 R /ID [ ] >> startxref 426475 %%EOF gpml-matlab-v3.1-2010-09-27/doc/f6.gif000644 000765 000024 00000074517 11422240325 016640 0ustar00carlstaff000000 000000 GIF89aá‹÷ý )))85,,583338882821,0 w??DDDWWWIKUfff`eovvwPjl[\[˜¸¦..Øÿøÿ8ù44â8ÿXÿwýu.©OO¹qqôDDøVVã^^ÿww÷ww÷ffèaiÂ7^ÿxªjÿ˜ÿ¸þ”7¿ÿ?ÿØùúÜú)ÿ—Eþ™Qÿ VÿªFÿªfÿ³wþ¬rð’m‘ÿq–ÿyšÿ|þp¨þXÿôDÿõUóçLÿöeÿ÷iÿ÷vÿø{éÐh°¨C—¶22¬DD®UUµJJ±ee½hh¾bg¤TgÖÿö6ÿ79ÿ(/ý99ÅVÿ7Cÿvÿ5RêvvÄ~~ÇmmÊDPÿIUÿVYÿDLüU`ÿYdÿUpôfpÿitÿwwÿryüciúFFÑGWª©i«?ÿ¿_ÿŸ—ÿ¶ÿ4¢ÿw€ÿ|…üqóD®ÿG¦üTµÿY·ÿJ°ÿe½ÿi¾þz´æP”ßÖÿùú)ýØvÄÿ~ÇÿmÜùQÿô\ÿõJúòaÿõkþöqÿö{ÿ÷`à÷^á×I˜§µµµ¢œ¢Ì“’ú‘‘ÿºˆþ»»û±±û«ªÚ°³þ‚”ꎤÿŠ­ÿ–³üš»ÿ§¦å­•Ñ®ÿƘüÅÿù™ÿù”ÿø‰ÿÙºüͬÄÿ³Ëÿ¼Áÿ¯ÿûºþú«ÓÍŸŠŠÍ™™Ó””ѓ̪ªÚ°°Ü‰‘ÿ˜šÿ‰‹ü™ ÿ¿î»»â«±ÿ»»ÿ±¶ý¥¨÷…®ËˆÌÿ™Óÿ™Éü¼Áÿ·ÀøªÚÿ²×ú€ÿ÷‰ÿø“ÿø»âÿ ÿùªÿú²ÿúžØÐÌÌÌÝÝÝßÑÑÍÅÒüËËþÝÝîÕÐÓÿÆ×ÿËÛÿÑÏýÂÿãÌÿüÌÿìÝâÿÚÿýÝâàÄËËéÌÐÿÝÝ÷ÉÊÿÌéÿÃåÿÝàÿÑãøÁÿûÝñÿÐÿüîîîààéþîîöîíéÿãÿöîÿþîñÿìîîùîðÿîøÿàÿýøÿöðÿþÿÿÿàßî˱Ô!ù,á‹þû H° Áƒ*\È°¡Ã‡#JœH±¢Å‹3jÜȱ£Ç CŠI²¤É“(Sª\ɲ¥Ë—0cÊœI³¦Í›8sêÜɳ§ÏŸ@ƒ J´¨Ñ£H“*]Ê´©Ó§P£JJµªÕ«X³jÝʵ«×¯`Ã>}'€Ø³hÓŽEÀ¬Ú·pãú”PÖ­Ü»xó²¤ÀR[½€ îè—B¿¿ƒ+æ ïÁƒÓN{€àÁ;‹Ó<ˆx±çÏR)QÚ7 A—!Œ¦ø®@€Ë‡í ÜÚ%¾ÚùáŽhY ‚›ûQ >ñ€Èœe÷{`€ó ¢'@ ztÔ£[×^]z‚íÛ¿þw—ž]<ùQ¾ ?ž»ûëí”?ß;xûíÍë׿{þìD0Ýt ü—ß}ÓP{,°@vøÞ€ÞWÅt ÂÇuþ÷@“\x†&PA P€ªxà‹¶çß{3î—!ÞµX Š @õøœƒ P€ƒGY€¹Ü@ÄõÁ4ï8F‘PBPg%¥NùQHEηxT‰– …‘æFø ÁQ;~ôQÌmµ UЖ-ÐÀ „ÚÀ†":@—°ý6<<@‰iù%ŸÉA=ÝãÅ1!æSPÜóQ%•0ô„©ñ‹FþzÒÎEí rHX‹6T¢†öÚk®¢õSA”\lDûðÕÖ²Ìj?›ò„˜"‰º“9O̪ ù"ÅFddë¬QP `åÊP¾¶[¨º½ã%È™EIX‘%ŽõëX[ŽE¶”æ„OcŠô¬N´›à#—ÛYG”Ç°í†8Ì!8ˆÊ$ þßã ¶üÆ“'T„ ëkˆ _²‡šà£r{ ² \ÀŠWT¬ÈEòð$>„ B¨rDÁ>1aDŽÁˆ¢|+üYLÔ3€J$STˆ s¨Ã,^ñ‹&É_Lö×’šµ«z‰!‚3‰ !‰ ɇÏT²Hè &æ@Ãî8’8ÄZ ¥±èE™ò%¹È…K°¶ZH•+eB ‘càÁ!’<\JdÙ&¶kÐÙIò˜=bñ˜~ÄáeF–ð" «S£O´æHH"$—59„QòD4pÒ$ÄD(ùÈGR²$ÍT‰. Ù)Ðò'FŒñÒŽ=èr&ÉþP\J„Q†o3P[ è¯(Ђ.3&éD /0ÖNm %žÑ‚ ‚Ž=8ã&í€#IÔÁ†4Xœí¢@ *Ò?šòt(áLX’{@5aiB¤É^¨r!BËKd*_’ä•{XK*¨”t G è9I’Ð’ü‚¡"|‰B͇p!õÔ©DØfD#ì`‡n6⇒”ƒ Á€I8bÔ‘"Õ Z\êHš:’³ve„SŠ?Þ¹^ñš{ئCÜ F<â+;ž±ŒG4¢ÍpGF|’]<2&k5H JR.S r ]CrÖVÅ¥y .Ú{\5!ùøþÃ=Âg8v’];–¡ˆe\D£)GnŠÙjÇ èq—[ÐІä”&)mM* _H5'Ô=HR«ßEÒžÑÇ#ÑŒ±B$ŠˆÆB²ÛànÄ i˜hLΦ07¹ÈÝ"~ó«ç‚º$‘nMxÙ‘3ÁC'.HU¢Ù]“c a‡"¡‹ðƒ¨0Bì^Œ48Š4¡oBÚª\ƒæ´ÌD)i‘È•†î(Nxäkð.Äx„V+âŽE4C"¶È1\V“̤Š&Nî~•šâè²x+³ iQnæ`¸A!~#l«áäƒê…H ߻Ӫ5P÷(þ~•|\ÿ~d´¹Ç“µ"3¦„¡b !Ë }&`Hø¡ˆ+]Á‡ÈXäΕøÄ&frLü‘§é„§™ky‚i‚@t!n±Aˆ1惸ƒ¼#Á±A0MY‹œ5x7!j15@ëåÖúÖ¸îo“Ebi­¸¦L¡°¢Ž×d‰iÆ"Ì+Æ>¤Õ²™gbd‚,€Öȵu¶³mk]#TÅp"òUöW§´Ž!3Fˆ3ÊzM£ä‚nH²ÐiS»sµÖ/¶µ}ëƒÂÎYCZµ‚-<3… NÈ1®|ŠºèÍíCÂAq{P$Šp»+²†6ú¤Ú¹6®þ¹ÍíûºÙ#”ÞHöNäLB ŠËû!…˜'dZ{ˆ$Þ G4d®E(~àƒ ½èAÅ(¶‘füX ìÇYžOê*ß¹Þ7®OÞ‘”g„ïI‚#r&4%Áæ²A¼{$x:Bö±QxéFÏûÑ}Àw£‚ áG#ÂKtÌ[!ø0Õsrˆ<ë§5×9ðB+·HäS¶Ërtí ‡ð3A0ïƒØã?ß} tW„Ãâ Ç6D‘z£‡" YFéÕ]ê‡LËŸ=i|?컬¿ÖÇ—ü®9’‹°c%ÊQ¡ZCÒ=ê²äÔ/H8FþáƒÔ+]°gÈì»ÿWÜØç AÄE#Òë ŸøH~äÏ€åkä©[ ƒÚ•r烴 +×pRWèehó EçžÐ ˜á | áw§‡ðp7'n@ñ~¹V|¸¶“·^Gv%61„<5Bñ¡^p’dM¶uûÀ}A¡Àt±­`t Ð€[‚èí'²†7 Xk#¨•çåPöæ÷ 󰂃ý_MÑ p­P{··á@tCˆ]¥I˜g°æ~òQ¸‡È§|ߦž’…M1Bæpþ(©t e†1 zž  1~—‹i jC!| °HŠxŠñç€ò‡ñ{YAˆT!Ü5ÿ§ŸWÄ€qƒá D÷ Ý@t>h›È{DH!|@Š´æŒ©øŒQHŠS˜%k 9D1sÑ7MQI¾ÀW‰hwˆlÁ‹±ÈzóÐâ °ñø¢ž@„‘†Ôu³•Ö“VWC©x|¨èŒémUò 0@“!)#q‹çcgg’sHK«‚)ÀÐé‘ZP^à‘nð=B³‹iHÜP{Áˆõ€ H@þD€ð(1Iª ±D× [²T‡ Á ·”"æ„YJ‹²¢A³¦r””nôV”WaFé³Ð…ázVc]&Ï€~á Að?x 2™ ß°°€C á |‡{ñ »GB)Yà) C€wg~¡ËÀRµ³epMÑ„1Š÷Ù¢ø)Z2,!w2J&̩䈮„P0‹åà ýaAh浓ø–Xt©¡ F †¬g´iIÞã¢è¢-ºŠ I/œÑ•àÀ!™‘éb°Oq âHDÙ6w½·‡¤ÝÇ–ñ C ‘ QZþóà@@šHt‚xƶ[Ê¥öY¦‘ÄU\ˆ.¨rf΀hñÏŽíH1yž­I@|*ÇHî©@úô ©õ)©! ñ Χì•©V!1³›ôvª£:è§*â@€Ê¬° ™°;€;p­ôÀõ0ª*¡:Ë°H}Àl!‘PA•a¤È-J¯óŠ˜“v£q‚MÁKR` XñÞ²óTD*³Iª|·¤á§NŠà 8€:` «à ± «°<¦ —HPA E(«p‰«Ÿ4¯Ðþ².»öê²ôʼz¦ú:Š–›IA¬X¡yZIÅG§FCç Ûn9®;`ÝÚ¦;ð­AÛÛ`ê[á ½‡³ÎÙ¹Ú²0˲2›¶5[’ÙOVqN†WA «n0pÑuÈæÅ}@´1‰ô°7p ± 6`¸ñ G{©G KÐaGRQ¶3›¶1³k{%7ÛîZïà^iŸ˜VºM‰u– K¨‰Ûº {DPF¹a6WºpSñ~ikŠ™«¹/ ¾Ú>Rá?#TºU1¶°ø8aI¤¹Ìþ0é£@F€ á 6À A6P»q;ꉮìà?¶i^ ³D¡¢šÅ`¿ö{¿÷ë²°¿÷Û¹p6p˼Tp 9¡·£§cõ†A@²ѽ±­V;¾6€±‘—è:–µjºðWP¼ûË¿-Û¿2«Â,pv0 /ÅP`êÀN~yOx:žðq(¯[à+¾á 8`ûÀÁ? —ý¨Ñ;v0U¼(¬Âù›ÂøK³öwÙh<[™<ÂDØ ¹^Û´×Ä¡ ª`¦€Ì;¾1Nœ’¼{l÷»þŸ¤¿[<È'œÂ«h¼Ì9§TA aœñ”]Âpyý I ̺>PŒ‚ Ç: 6`J,Fçă3ɶdô{dZŒÅ¬,ÈöÛÂQ‚ ÆcL])¼D4vë'[óГ}z»Á;À~‚ œ  ½‘ ™Px<Fwªîà[°¡Mð+¿óùIšûÍüû¿]Lº\šÖæc ± —Ú΄ÐÆåúÃ~Ê´Ë›€ &`) )&P€ ñ°ñ(Ð ý= ÐEwŽ»ÛKÁÎWñ~ú«Âý›Ñ-È° MEH±X·\1 ›f®•ºñþe„?­C žŠ›˜0Л@Ó%ÀÌ qÄ‘­‰‰>0Ì qé QT|÷à ¾@ úw ¾à£/aÑÑù{ÑÝÑ‘P;³§WqºkçOCKº'ö`¡A“QÇ QÓ)PÐ !m²±ÉéèÓi9ÛÙ^ëðbº°¦ a®ôîD JíÔ] PÀ…—™¢(ÈV=Õ}RehS±£G¸´°i 4Ë£GYîÀÇó<°0¨a ÇŒ×ÀÏpíšP ±²+õðÉh)Áü€~‘±À ýà`qaðnòØAØR³€Ü$‘ÊÖFÙþ¿Õ‘N·šÙnjºÈºË½Ü‹`;¾­¶-®± v}úL×Á )€«`ÇI¼ÒÚ‹°rWÊ¿=(ô{…¸Sp`,ÙSÅ­ÑÖÝÍÔ$üf_¨7¼ÀIèt³¡–± Ä0Ýà;Ç!ül% Á <0·=C¢Ž q ”ßѨ áËýß/7€³¾Ñ п?ÞA~ÕöÚRq‘`ÖQÆ!+)¨­Ð}~‚ o<ä½×P×p×€ ±±Q‘zz™¨ù042Ø‘yP=·à8N=îã”ä ÎÐE}Ñçþ{f¦M!lQ3§¬¨ Ú«Û@½¤áZ7`¾ÑÚìm&ðå« Ä‰A@×<½ð3ùí)­^mïàN)¯a@äC.ä®äpç qJ–:ïp]Q‘N¡äR“ÝX¡ÛYÚÁª‘­‘å¯})€ ßÜ+Út ’õÛwðÛ»° À€Ü —i]q¾lþO ¡êÒÝêÕ½k”9=pªÁð× G/.Û@tz ¥A1p±½ì\þÞÁ DܱиjìÛÛŒ Àð | í»P¿ îQ ç&‡XÒ!!Õ Àê¯îê"/þë áAÓ3cëä!^á õ ÄãJ»% ðAðAÛmÜš©g‰ú`–üñÀòIšm,E’îÒÝ¿‡üÔ£ÀRµ¬©™9ÈKèAaÒ^t°— !;`^ò åáó‘Þý@Ì+ù´áU¿>ÀÉ Ñô'!°õ ±!?øB>ø†ø&ŸLcäP a Áö÷åL~ K2/4¡‹nå6âô`[Þö÷ÄÜž¨ÅP È –›|¸Ð=ò…ûCNø?‰vƒ6ç|J>ƒž·…žaá‹tþ_ìÕ*Q[˜àåáöÑÉ {öCߣͻ¹ó Z ±åöñQ¯û¹øëÿû!TR1BÎ}¾Ðõ¡ô|†ùÛyw _A¤ßÀ~Þp$ˆÍB† >x Ó>„ÞvDò­ß¶  (BŠ6°]„fÖ@$âJ–-Þ{âÏå̆ f‘SçÎ@øì0€æP¢'p)šTéR„0É1…UjC)2ÞCð „Ø ÔäÅ~H`!Ä®¡ lSbâÄp‡7‚CŽò!j #‚€ÚÄg"Ì{n—šƒb˜¡M— ~þôù(ÏœB+@åòæ¥P|¹þ¤ÆYôRRVæùÅÐ/A~‹z Òj`=W]dÈ)…hñÚ KU?{A|„ë÷¬™H®ýBã9ÊÓÑ4i…á¼à&å:»ïün¹ºK ÆŸfÖÌwèÝC CÝá¾3 óíA¸ìk?m>|Ø Tº†â)AÑJ`È"‚ˆ¨§ŸW‚b Eúú«ŸwÔÉ*=ZÞkIŠÏ.Cà¦î¾ãŽ§îı! ÌkñyÆ3Rœ Ìc °Ž¢ÔÂG!„ˆR‘Çëúq]l#þÙF?ô3Ck ÂF0r .~¯*É3Q^PA1Þ¬;/ pGí0ÄpË;Þ(€;ö @0=þøŽ³˜}î²~˜¡s|‚cgŸÁ¦’ã‰õ8,Åý¦~ }s÷eîá„•tÉÎÄ@HòPmÄùZXŽ š(œzEbà£×cû`pÓíñÄÒªìGtÍÅð ÆÙuZUù×A>v÷”s¾(eÚ ÓjˆÀþ!R?ý±¸Á¾Æõ{ «Ð{C<l”Øà@ D °@3è ¸Šy<‘Ò*³tqâ„É‹×â45BðíÌ«ÒúaçD€»O*«")°±WÃ…= „À°{À”ÀñP€ïsB¨ ¯•ÀƒÔ@ˆ‘`ˆcÁ5Ù²ßX ‚{Øâ'L°½2Ô„U°ÖB$Ø9(‡}PàŠ6>¡ŠMs·{¸¶¤Ä~0#7(Ä^|C¬ŽDô•ëhÄ©`•†ø„hh ˆÝ ‚}C¤†ø¡‡pÀ„¶hØùK¸hˆƒÓF£"°nà-âÓ9ÌAˆr <M;Fu±¼msÀ$DtÀ`þ¼ ¤œµv‹ªwˆFl YDˆtØ‚ KÄ‹e ¼»–%ãC4$:†X…HÁH:`" ÙwpGä#y¤Ç~°Ç†0HñjBš€¬wéG_‚@ƒ«øˆf8¿}ð Áj% OºIÑy$U ÄK(zˆ”ÉAá:‡yl|¡œ£0PHè» _ÔɹœË}”Ka b¼x`»{d §X‰\p;®½@·ÚB.tˆLП¤¸Ê¢ãP›}@q‚áà†¶WhfpNéÀ:ì]`H„[ðËþJ¥ÈÅ]ÜÇ”K»¤K¼ ½´)(ÊÔìž'ÈÍ~(‡2hg ›`†[ëÿéª~X?‡ ý ³¢ˆ}Øp?…Ê"0 éˆm¯¸A‡Þ!ˆK3—' ÅFô…òœ‰Ö,Âõ\ÏׄMÚä Û|“ ÓM¢LˆpIûPÊ~Ш “AˆT ‡¨?—ÀHPPPPXLhИ@Þ{PÉýH¥\ƒ6`ˆ}¸OÝ$‡[$ŠÖ”Mö”M»„ÏÍðÉ€¡êB̹(lŠcˆ>X†X}ð ÿÁDŒð‡`…ØÛD%#"(ÓE2¸#0Ê|þ…Ò ‚Wˆ‡g¸·~ @¢{0]hr`¾–¨Ñ¨ =¢Ð8ˆØE¹dÓØ„Í|ÑËO¹°úŠ”èÍÀ@J-Mâ4Îõñ„}SLˆ(Å–ð/‚ð°h´Ù8‚!8‹pÈÐz*¾e¨¹û@„–,ƒ=½…1±S‚À¸Ë6mÓðESEÕ8UŒ9­‘SúT— …E\‰ŒÚÓbxH‚pÿäny…†è‰LpRžã„SÐà‘ Ð!}À ÿ…}ˆFgÏø»iˆ08S;EÑ?TT-U0UR…U]³U““:}Õ•¸‡)¸ü<ʆ€6†x¢)¼È”Iþùª•`…éÄ–UØUøWˆ`…FƒÎT@ÇÀ§‚Åk„eH™CàÒÀ‡5µl×w³Q=Us%W¹4Utu‹Ý•Õ¬Ž™|0‰”¨…zˆü–á´ãsŽþÉ5„°¤™°ˆZ‚¥‹Cý9…Bf Ø`Ÿq@±Üv؃ÇóM2Ø2„ðÞ$ –å¬nµI-WqYrýZ“ ”½=:¶¬u 1XD‚ˆ\0…ÜÒ†pã|¼hŸ†Ш Úè# ø ‡P ÒV؇|H Â>À³6H|/(¬¡P[ÂM¥¸IÀ [R-WR%[7R×0A'þÝdÛ•`Bñ[‡Xå#ˆzeG8¿ÿŒ4‡0‹¢8ÐE0à˘ÂÕ!Þ„Iƒ0ÑÄ4ˆ¼—€‚oíX•Õ/ Y‘%Ù»LÂѵ‘ç½Ó}·­ÜÕEˆß”DՀǮ$—퇆ۡ‡@øa MØ" Tpnˆ ™VAé‡C؃OÉÖ=¥êX‚^pý¢`àvàvàÐ V}úôKïe:!@ñ^@ƒñ;½¨mˆGp6¶Mp˜ð$QÊà' À¤¨b<ò…Bž‰4 'vä&þÚ`¨ `÷Õ´¾Â+-Ðä6ÐZ O.Ób`àbZƒ-Ð…—D„?\w`„ÙmƒÄÛÌ•°L^[‰oàJfáM`…)‹xq‡Gh„Æûß>h¶å6pˆ0Õ"~ˆðaÍ›pà‘­f%n`‘{“J>s€ÞFô^™qKˆ@ r0ƒ]pk•Z@iÉÕÿñºc‰¤³e„úþE‚EˆpøJ±x…è|„•|6îŃn\pE†f‚¿é=†äfbrÕæ0áæñè4;=]g®Ü‡ø3P˜†È‡C8É͇F˜ÆY¶`ã–HºX€…ÜR…£5#H‚Ƈö؇n€uÔŠàbÞ~°Ö>5/P_‚Ø"1xˆvèon‰Öd`h`«–è«žd¦0Û7©É‚ Uh†F^ðtnˆµB„ˆ†E8>sh7pƒRà­Oмrƒ+nˆzPᾎá†d˜R*…ŠÑ‡eh„¢F‡>8i¦F/PHgîM¥ÐÚ¨z表¥ˆÎê°j¬ÆjI†âþ€¹dS 7€‚1`KÆî‡vðƒÆF=nˆpðÿ)…ƹ\šR y…;Ò‡EPÿÝŠ•‘°2`uÀ`Zp´» 𺇧ÆrÍ¥æÍæìªÖî­^ŠŒµÓYp‚#fˆŒEhñ-é¶WfPcÆ·üÕ×V¸â´b m… 툠}è¢æ‡Cp\¦ö‚-0ƒ¼æd¼z'èK÷`mŠWˆ~—ìÖn ÏjîV ï Zpî{Äò”Y‡À‡<@ƒ/_ëøZ†EðßAzÖ¼+~Ì™ØmðÿøO‚h‡þž·?8q`ð^@¶Êm‘¯Àiþþ"ÏÎnÏþl ·ð¤ÀpÎ0û¾wéþ¤@vˆb؃Â\JE˜]âxÖ …Š)è²JŠ:nãx…z€[Ñq„÷òÂ4‡7Àzl_p?4-O $ÿ©f‰ÍUÓ ßl&ò¢èjùjÑ8dŸ…ñf u(ƒ°„[kþgˆ6(xƒ¡øu¥Ø÷¦¹l$>“c'öcord·êeŠf¿ %o„82(_㈧£XqñiÞ2ó¨t‰o±¡§ø¦žgPbÙ|¨øOæu>GM½&gOŠNÖ«Nþh'ô•hÍçì÷y0¡wHk£w€€ Æ(~Gç,nÕr|ÀL?=dz‹W„Œ¤‡u8AíDÝEû ´p…§ÜP„ÈUpY]bèƒ'ª§õ'ƒ|M}§y¨è÷¤ù÷ô¼‰ÎþyOöŸ®êJ#8 0ÎhôÙùͨl©¸Qô$að‚ª%ˆdøƒu~›þ‰¥ºphq`öú†@¤#8Å÷]…±y³j÷ø}@ìÆC -L©'&ˆÒ˜Q™‡0Ýû•˜ÉߘDö¿'x'üÀ§˘¥Èì~x ˆþêhü¹…ÇnѸÚÞıY³¨óÍ*þ|øn"ývˆ‹‚›Îi™†…oˆ¥ó¹¢•ãH{\n„ø…;Ðv`ȪuˆYˆ(äúlCð „mjicÆ̈æRLøŽ"0`7rìèñ£G\¸@VL@²€ Kh°¤1£¥K–K–8 €àƒƒ9 Ô’BJ0ô(R‚÷Â$mêT(-(÷:æB3a;@€Ú!Ôþ×È¿ƒÝ<ù(j[EqIˆQõ­#*T(6Îõ#Ȩ„Í…=Hìί¾ýÌm1H‘Ü“Y Ìø´¢ãÇ’›†ñÕÔ$H07s†y³ßƒwûú!@@çdŽES³îÇ´5ì¡R^o,‡&Eb}Ðu}ôHÁ}Ú<ý(;*ÅXH† ‰E’ž‘¹QŒ ÄB~‹š!ÌׇXB^dÒT óÄœ®ˆ Õ·¹Ê±±ÆÖ‘cÓOXù2IÍ÷·Ðo%>MBLÓO%FÕ×Ïj Ræ „Ýó-ñR†{µó]wÌ,âAö´rW@ŒbOB±Ñ–8H!&+¬ (,QVþ5bÓ;Š€xP2{ðv:fàÑaùE…%I|°Í!laL•féÇß~þõóÎ p`?§Mƒ@˜-‚T å„›‘™9N ´Ñ=i™Ð!{puP3Œ°\+@”õC+£¤â¨ÔÓÔ4IìpPd®áÃ(ùÊB€’Ph”SÐAø@!Å•N2ÖÏ”HͧƒO8U@~\òç%¬5˜«GRðÚÚ-O¨YQ9dŒŠ:$ƒ;pG6wýJ¢Õ£ÊI¼õ˜*°PªÑ<>à%Ø3Žt×Dz¥¦ÛA†ùòÄïyŒzföÐBñýºoE²â—™­]òKTþ›'…ÁOÕâ+G¹”OBºiJ?@2±6@\úÃ+ŠöËŠÎ¥–D‹e´Y@ Í#X ‰P±ìµÑN¸Ñ†¾ª ŒýœsNÂADRYz¤Y -%ÍÒÒájð®B»¯Ô ÑÂðFk°{wà #?Ί´Õ” H”ÌÚeÔŠŸ¤p;Š ë'B¼¤‘êAfP¸Q†HÅsÕBû‹e­M×°´âI;þôÀlÞÏ-MÆId³qt¿$äÌËåóÈ2šîóŠux¡Øϵ>ÊÑ*±¯b +Þ€De´Ë0eI³‚F=NÜ:{ ßÏ{ÚçÄ°=9þØà=o”³G˜OîÂM•°ãJÓÐ=gó+yáþ`ífóII¡XmÆòÈðÐ0 AÝHÊ<Õ³œÈ­²ƒ l ð v¬Ð vŽŽ¼#wºÛÅ'âŒKCŠL1þ |xA¹Gfñ¼Å8…p!GÒw=¡øƒ}ˆÉ¼Ç4~¯†âÛWÔ¤–½RD –£ˆ0¼1®í¡á‡# "²áÚ8¶¼Ez˜8È„í¢ M$Ää,p}H£,¯ë‡"€Cg!øp_B3¶ë)&ÔéÃ_¸°h Ù‡fø=Æy¯i7üU…F‹òþðLâˆ.Öð¹Ðõ£nƒê‡+ŠƒE¥¢Q9Å p` zPd‹ÁÄV±‘n!Äx9È2èGâ ¯^8B¢r%:’j(ì1CCÔãÃèÂ`­r-,ó/-Á¤i2t&Ôµ‚IÍ‘yT ¼&I„ÃkÆû16ëx9ýGÚ(Ž’#¦L èaRR„i…uVe¤Æ;H9¶€Ëƒ„hÔ#aSŽYGÁ L¡õ ÕŽb´Ž`q½(F1:EŽš:ä„^E’w„#ëJˆ>E1gA w©V=Ž@„µ”:ÈbB⃼!sù_EŠÀµPþ± ×úð§~pp ýpd-:’õ3"íˆCÊ˃ˆTJW­OžB«Ì\t5«EÉ«D Íç+Ü„”Y‘5ð¢;€0"Aê&Ëê”Èœ”âALaƒvn¤§==ˆtl@‘oá |Uf#–JIÎ9uBýÈV?2s°'˜Á,毲›¾ð# p\FQ{VµæŠ­B“ÂeƒfÙpЃDtÆAz{€âRcŽÀÑU$Ð#Ø0 j„ä*·Fý˜[?t@·§·¥ H3˜q»– ŽÊæªz”¬&³-lHëGî¥6µƒd-¬\4ÄpT¶à¥ÈZ¼!õãþœâ†&±ÓX a[ñv@Ohâ<åbBR0Pd}Á'ÜÐ~ØmmŠŠW¼‹¯xA¾ Rƪ¥È`q•¶ô¥3mi»IÇ×»‡"¤ÂŠ æÏyc‘g9g~<‚;û¨ŽD±}|cýþëqYÁy¤¹*Є ú Lh"%à„²‚`ôC±QE*~ R0™gĈ„îþ€[‚D2! Ž-I´PbÖ”÷#£^h _ÜZDÕ)åÈ*­i‹Î›Ò7š|…6!þ*XüV ‘¤‰åCƒÚ(ÄÕŠ³áŒôH0•L”@V®–›Üår Ø Yv?n¶ýç¥H g?’¡Ò~ØÆÛ%uÊÆšr?…ÏXU7k ú”woÄò¶´½.¡sšJù9ÔË«…qäÇà¦7)æ£~„£·>xb?ðd‚W•Á† J€‰éô£§O>@.òƒðàɨþÂhøð¬|ìHdâAò,”D›!#}w±d@ksyÙç©QºŒÉ ô ÏûçEÐÑã8iŽ¶·ÔW7é6õm$üÈ©‡­¡¼Sä*PÁŽèRv.ÎÅôcW¶F¤{´È ïG> qÝœš êHˆÚ¾ûÝņ‡Òc?½ãgî1<¯ˆÏ5-ôL_úÞAótg1…O¦«‘‚-šn†Q¦}Ð<ˆ´QPä¯ÖöÆ ¼ŽxT<õ™‹)¥Ó¨‡ ,0ƒ@DøC„,ø(ŒÆ#èA"„¨$Ä,ô‘G˜Z/)_xu„÷¡[B1ßc ÞPHÚÑôþ-^Ð=„DžÔ¼‹Ò‡?Ü©d^^)‚¦¨Ž, ÁWÄ ›8X)}YÆ1—¼^ !AüC0d‚ü¡BöƒÆ,œ;4‚AÐÝAƒòÄÕ·uV y I´ àˆ¡zIÜ÷ÁÆ£9_~ Î!ÒaäOÂTüÞ5Ä Æ „Ô´`~µ‹ VR8iƒu„Âh ‚œQ&è@ÄaC @HàëD $«ÃÚÄ ?(¢ H#¸ÃD¼ÞõÃ’d óÈ\Õ` ¿Ì‚\-“G@îbÀê@ õ!AL¤ÁÆ=Tm‘ÁÓþ|„S7A8"”é@&œ’ 0á%öÃ5\œ&rbF¬­%Á\¸’,ØÃ>­ÑʵA3bSžáÑbœÐ㯠 f/îãС ƒH A+^S-8Á->ÆÕÂ$ÚyW?ƒ Ø„E´ü@^ôC:„<ÉAă ¤€`ÀSE[—‘ä¥&ìÔC'Ü@ô@ô€ô€u¨ HB_üÁăô€[GðÍ!„ÂæX =òÊá8Åö|„ðã>nÔõ)Ová5-0dR¼Ã;Є_Gä×TµÔA0é#¬0èàE*@™ àA\C p#EèþŸþ™ä™@—™@JRJ,ë&¨€Æ,&Hƒ4hÂ&‚2ôÞñL`R%áQÉEÐQRÂJ2JÆóQ@%/úã‚Dxx1>UD-Ä£G¨Ã8#UD!-DÝP$5öƒ¤ÂAxC\&'”€ä;öá=D“°&Bìç©ŒÄG‹~ýSÔMæ¦[öæ¼Ô5þHƒ,)Aƒ†}—"D’ðÒ‰eædn„òÐ}ä£Sºg{Âçzæh|î(}Ò[áC¬AŠV†-Ä9ÜB@Á#‘D±´"-mZjJ7XG[¾å5(B¨@„z Ä„Þer„_&„³eoâÓìƒ50A¨\ï%„ÞyÄŠ¾(yžæF€`â}„.î¨ÊçæhÆö!Ä1”ð!žöƒ?ÜÂDÁDêD„Á-DÔ/˜”Ι;Bþ€E„¨ƒF—rÄ5”¤&°À&l‘l‚ýç˜J('fB6²NÛØCq´BÐUB œè(Â%èÂÔn®˜È&ESæ¢ïò®Wq/IPƒ€4¯ vÄ›ÁhðãAßRwIÝ Ìöæ8 Ç耮m)Ð"gGŒo?lÙ]C°IÇ ÄÄBôÜü€:òS?h?+BHV.kfýÌ{8qPð¢ÒÜÆ-HŒ¬ïþ®Ž:0ÿ®Ä4xq<@|ñ«Œ±Ä^ð‚Ëã1#Ex^ý#\ ¤€¸AŽôÀ ôÀþëé€)DÐÖjGx Tî—YèþíÔ\4®ÄèÕCÕú€+ä§à/A”C D-ŒAòeyaò{4ñòâãdä.B€(‹2ïÒó£,•´rP€Â2Ià/ äB-_ÏáËî×AXˆ œü@œÓ,Øè%Ä•©ÞF\‚Š‚ét¬0 ì`‚àÁ&ܺB øÓAxÁþ„9,eSüïÍe2Œ>ÅØ®mÙ.jïéѤ)u“rêdq„Pì„NôÄATðdäÂ]]¬Ðü”ß ,UÝpÂ(¼C ÐC–t% 6°*_–$Žþ@G“éש@ §À鉀 °Lö€ø¨£¹4r:H%IV&gEÄ€'M‹MãÑði¬‹™í5™kL+GÈ3P?ðP“r§Î)w„€%p”%@›tSàƒU£ÁmP5¯˜C@ò¿åÃ!B’ èôÖäOèðA­öý 'A¬uI¶u—Ù%ºéqrz€q·bi‚Á>äìupØhƒJƒ$†dN:¿ØxÎÜ¿-kÀóAhvê u?°µì˜ˆÉ«Pš<@¦z€HM4¥jK¦A´/du„”C@¦lÏàÖ‰>î6XÞ þØ*Äøîs³0Hè‹s{'Ä9ƒ üÃ-, Ã'Á'àC,õÆô>ƒ†õï̱!GLIbGDa°‡²¶me¾¬;¿3I @QA u“7°|ÿG€ôÀ TÂcÀiP„–Ä€³†:üBxž¨­¢LðS°ö 'E—sù–ƒFP€®ÏÉ;ü¯ûúAô:þW;°„°ĨÃ1¸A¤äÁð‚¢“C±IµÃ-ð‚ ¬.˜Ã±S9WaèÂAÈC<‚vMâŠê8È„Áƒ f°¢&¨@µ; üú°ûq´±Wƒ<èÀ_¯ƒžP›´Â:0BXƒ¹ë‚hÁÍhÁ8ÜŒ¢¿#;±D°"‰âÈ[»c|ü>„ºµ‹¼a»Aœ|¸¼¸ïÃÌ|ȃ<Àÿ{Ëë|Ï‹ûÎû<Î}?X‚=Í<%H€®cöF0€X€ ¬zQ»@QXÀg(µR3–ëzSÔzB€½S'Œ¢ëBàþk+¹)Dõ«´‡k mÈ!>wÐC8H ýê&*(ŠìÇ~ìÏwhÄ«SÄÖ؛ƭùoDV(½Ü 0,„ XGo“*–÷šo¤ؤrq"ˆ…—¡€$úŠ€MbB»^Ë' 3p?ÌÙÝÇÀ l˜è/_G°|,Žúï#EO¯—Sʾþ˾—È>T9@ôá]¿J&T¸a•N€ ¡ 1fÔ¸þ‘cG¶´ÇÙÂ|Ì–õkÓ¯•  ì 4¢J¡S M˜øˆ $C˜¢è§ƒ•ÀTHú $»Fû ˆç—B'oz‚4L#0®µh\©plV³gÑ~Ä%mÆ ´;—n]ºêõ{÷à‚iô³´àA´XZh ÀEÛF–Y$ v Ù1b'Ш>FíëWÏ*…7N1LIèd&€µ)ð‘~Ú~ÑÆ¿~ˆˆ ö¦Mñ6p¬h1ÞFWÚ`Á¼Ó²+ºW…Ö[ÞfËö¯ýÎlW¾ÜœkóçÖš…þ­GváÛ š#}´”ÌÄ´HLÈ'·Púyd³vöȧulLÈŸ'Þiã»''Û*­(IÕ¬F•Ó5¶Øƒ‹:Z •ÖZmU.[É”sÚ|€˜²È\þM#:û(FSŒÜi3[òá‡y²C ’çР¬Q¼±¡5 íù ”}–y¦Ÿ|ö¸¬4ŽIèž'ÊSÕ5SÑBu,%Zµ'7”uœ0l¯£¸l­uW‡æ5Yj$V ¨YÙ‚“uæ:ÕHŸGÑ'¡yBa1Ú¶tÃ…LÐ$¡ØZ4+y`S¨\z„2"–}@±Öžgª¤~rÉE!)p Ê}[íŠ!3X`V˜ãŽfñ%á ˆWˆÉ¦µ×­m*ÛIæã?ˆIPd’õÉX ¥|•„P!‚žš2Áhš7²™Ã…@Œ§™Åáï~ˆHhþ<™ÇE:$¸~~AC¡ZÆ°[ßR£^’¡°žÆèj¨YuõµŠbÀ¶,ûa±%†]!µuÌ™bFêãdäå(ÚGLÖ«Ÿp<±”p2)n}"¼ðëjÍÐkt¶áñ#L…Âm.÷­˜Ch$C¡[¤P¾ŸV_תtÕ§®:#'Ñ •þÞ3ò§})âÜábt8[ï6Æ¿Žäãw‡ØC†WîðØÈ&„KÝÔš: Í Ä:ëÌJ<åÙzÖ“#¨Mm,W‹¶Ö¶Åm{D1Þ( |4ÃŽhÆmbVB¼U=ˆ0=…°âÌÄ ¸fnÄ%àD4ûá0SC°+4WøF˜J¶+_¬7!ã—²Æû>®„÷Tû“ç-ÖSÚ-Ña<öqücÇwkýŒLþÐbäW@îX†"ükÈ…„C)B+z;#AÀݤ,œUófáÄD”S<’e°pCn~°=½‹$áÅbD".äH ÁŒ?þ-Dg»8•¨¿Êó¦U  =ƒi!GšÈ3òY˜²2ñƒ¶Š€„Y5¢PlrÕ5S[‰ O&d6p.F4Qîy¤£‘‡ R`3 ”eôÇ'¼Ô Ó9AwÎóö ¥0ËX#I e ½2h¯úÐZõw˸J-~ä“ö¶¤ °ì­Í·#"ùƒ¦«äH(âÍwBœÖ „z IÅo«ƒ÷­Éîä[âûÞ+]°KoÂ{Ö€ÄJVÄcd9_ÊnëýøÈc„=¸Aèéj‚ü[†9GMUøB®0×6鬉î“F1Ú Lô þî¡ã®‡ýØ;ò†eßží®* ÿ]C§t¯L®L¤Š#`p°0ø”æÌ( O2ÔmŽŽ#ìaL¯ñ– ò$ï¢b (2‚†Ð¯û˜­f6B¤.l!*σ*Ê ªe]Ü®AÒ!œúÁØâðOÂLT¦G®ÿîü!Nð#Tΰ!0Y$¢b@ök–aù˜Ïù ª°#Taú0£La–í´oóøšÆl2/¸‚¢2¤ ¬ä4átPú̳žÀk<‚`ªfã¶#îŒÃî"CIˆðÑ㈔۞00 ]…ÜèosÐ#a¢þ$âÍ%<#ÚJ ‚bÀò²Ïß8DÖ#¶§é$¬Õ˜ d¡È€-H!îÁ ¬ž%Âγí#\‡q@–‘%cnA äd ¢ %Ñ )QN†/! $]C!áÝ‚Ê(Ä@Á 3Õb!#(o°O ìF 0AÖ¤4‚KÇpTà¯V('VHDDD”àŽ!w1"!îAþÀ“x“œ€²>bTçª #š=0²JâñëNNhÇ# àŸ0ãdûÐ=œ‚Š#¶aÔ6È%ÝÊ¢¨Otà{°K€ÌêQÂòJüþXhGâ>ýì±ÕøMŒ ì–b*¥‚þž…<çIîa"o&9b#Q!®m1N­½ a !.ß2.á.çÒ.Q²Ll)šÅ%±0&7¬r)Bá˜p2Õ2‚v„.D´éÂ5O%XÀ8ÊÀ ¶C—ÒN¤2Áqèƒh¢*s£ô Ý"t0"ÝàÄr2^gãñ,›hmn"{¯#öAä’.}Ó.Óò²JÈœ Ê—¤0Cm0ƒ`¾#ÄÁ“ú³1r"ÍB2 Ãro#®N°éx›dÂ4$¢ gÜ¥€'þ!"²ÜÀ nAÍi‰€Á  ˆ¨loTHŽ6B× ’­J¨Q!vó.sAT8± NË<øAZ¼q#¸¡y8(Ë":‰@P0"bÀ:"R`à.OÂl7Â[ƳzÀ"’úARòfòö@ ŠÁàóõ¬¢>¢_§#[ @Ûâ?ÍãXÌ@¹­7ïR.4.‡“J" @œA'CŽ@­#PÆO°L#PÍC¢Õ «N#EQpM E1JÂJcF?ãZnôV‚‚iŠ±Æ@bãДäT! mÐÛ8‹’å7"ër.ƒ8ßrJY"t2&þT+?bvhç4t!rR#Ú‘ß “+DʯM€…\(‘R#Vƒ!2Aî<å¢lT¡t4àÏ- Æ ´¢##9NÎ+2µ`’U2œ@YT!ÀI¥uZç€R2°«BÑ"O¶”#BBá91âÞB¸x|ò:U@´S#°sU[H¯6 5a€Î Æ`©ªƒXûÓX{bYIEžÎA©„Õ+fó<àa œµI§5cåÒZÑÃ"ø¡þ\ã`ò#ì!>Ã&7"l#v ¹6Bé®A9OÔ;1â T ®Á S V_µiðþ°eÚ`û €´a;¢A®ÙÌÎ8Ž5½´”jbÍC ÑUž5! . !cÃÖÆ6. @ -M¬Ú‚AS=Â>BL‘ º,#Ê%58ÏÂ2â#9DM³¢M'/£ú!‡"Ä *&¡hó4>ûA (ru Ñ ô“ëô!ú°Pm¬_zpmi f-]…Q5"Zå2lU·l¥ud–c,`Z22ÜÁ›6µÊ„6eñúΕTâÖR`g"-·SpA‚2¯ N Pá"Ä*¯¥R·Wb²­iG…x™6~–a%C{‹Ô{Ïâ` ¦kâXWc{³þ8Ö<èá m¢ô–b­4B ®0¢ÕDÔ„-¸3à %°j‚û¡6z±4µø¢¾’»´Mt7BH¯×tz≌#ŠŠƒsug%Ä×,BòbIlQ8…ËvlÛ×<$`,ð‡>¢%ˆ)ú0òt#Fµo!ì#²áHàäA4Nõ#…;O×8E&éOìðÌHa“b,Q/˜Æ˜„{‘•Em–ô=}ɶV—lËØŒÕØàAiW´#XJ¢6 ŽàÝV!D_¶¸¯'ÞF`H ªáF º³#HD2Ë»Ìý~`ohK Zþ/“¨¸È!Œ±Ö?)ƒ;W#gom6ù‹5ÇLØžÖøŒS͸…]C%?¢v_AYj6âƒ~.#3]Sèó<â/:/¡[$ˆn(û! ×,jêЇ¢ ʼn jêân–UFå/8¸8B«kõ5ˆr°`0/ï 5cêàsGâ¸J¶X2bsjiÛ#ìº'b[Œ;‚* &•SYu'á·+ û±ûXþb¹–7âRÂ#X*²ox*+[¸L ;™û@7#´š;Sþ!6Áeq¨™ñèú¡v)9»‚SÖˆ‚˜€5b.a˜a½™â»­’Rú’'^‚ Ô¡H1NYÅ+£Öµ™\ƒÓ‹t£=§§}Á­èÂâ”# µówÞMÄQ½EéJ8;‘ÂV2 …Æ…x¢†b„ÎoúÁ–‚ÏutvÙ *Zl ½Lt*·œ¿þ]œËykž vœ!$½A‹þ‹þè«]2®ý½uÔ—" 3‚ b´[!Tã1"y2³"V¼X#Âèœòâ(¨27:(Z(™×7e‚ vû<¶ùìøÙuž!†Øg‹ùÇà~Œ áèÍxð‘žèŸ–>2š¾dt37Da#|F#®IÍ{˜!:!V€UõêòU‘Å…º[!ÐEî&ƒdx|ü*S`¬ã÷smŸhþýïAáÕmFà•nº}åU‚ zõ€J g”MB½òƒ\ñMG,u¤BbBu2E×EâA"ÄOI™rŠABöƒóôóˆqýdfP.¼x†ƒBÕGß~øhŽñIq‹‘B (˜à –¥¤Qå$Fñƒ+‰cDG«ì×^FísEÒIEÝB'ö£C“£”^A}¬WAqZ•A™ß‚*•ˆRÄdþP.ˆ`”‘2¸(Ej ‰;BE„=!‹†-M¤¢QmJ5š%š¨IA¥d ^!d>øàé3—ݨYA¼ìx9Uš*E„¢LÊ*«Å²ÊêRéBÅFÛÏgÔT€VJ9)” ^ÛÏ•@IÔ(>ˆ²ÜxYæQÅ7TÅy sQ'Mý»Ÿýxñþ€{бM)e(Áקp£@=:i&øíµâRäÎ2Ç ÄMGªSŒA¿k9=.¾î¿ Þ;Âý/Zý; þ–•!OAË›ZÕ¢uµ…èCcA±Ç­‡:4œƒ|±¿Äe b^h'>°RŒXÄr?¢;RD#òç R¢7µMqQU4È2Ø!”V˜§#ß@BGt0ª‡¤`eFahöA‚ïuDé3ÈVᲺô#F^Ú—Aˆ7_á¯?Ã2\"õÓgÍg‡Td||ÑKG6)ŠQDž™ÇLÈLr¡`PD„þVt$ ›ˆG•´%U)ëHJŽDŽÑF~ðGvXsv:qAnÁ´EM 5E!ÚD~Y¥} ê†FÌÖ€ ÙLf*Ô™v˜$¢*Y£È„ AB†BHEî¢Î’N,àhŒ,er'ɯèèUAÖðÙ3*ý” @gªa)Rq9@4Ÿ™P…BÍ¡‡‚h?¦”äPækbúæD<À ² J6F@B©–­ ¬ “AâŽäÌHH: ¢‹]$ 8œ?rÓ¶¢ 3"O'ò(¡5¡QkhÒŒJ. dF“W)²Uë,$;5Š)8z?2¡î~ðÚ 6ˆ#¢ZCäD'ÊxÖͼõø8¡‘0øši>vš›ý*”x.ÿK,ÄÚ‰°P@áºmf"] JH÷ÐuˆƒwÑE‘óBy^O AC²œÕ,., Zþˆúnâ¥â{àÐÂð׊ÍÕœb™³ØÅFzG._€àòç¼Aâ;‘ò@y"âàíDèÑzŠ p"vŸGŒþ}À]ï7îÇ¢áYghíÑ)Q4MÀß«†ÄÀYüh¿na‡|3§|X“P Q šW1x@)ÅT¢a$›PC7R96+µR ·}³×ÅPlq fm€ú'ƒñA ÄóKT_p|2§ƒ:(s¨Ó'a Ð 1†EáNuô54™ gaÊe˜ ƒvx·° .ÜðN¤i×~0x‘XF¸$þ3Ȇµ87h @s=ȃÇ÷@x—<—gFlB1_>p` ±`Áq%pnñáá¦Þ&GNˆR–4eT¶xPL0Ximȉga-¨€!‡9èƒ98‡txsð °O–°@ öA @a}( F‘CÌÑ-#«˜Pö÷Kg+?@X¶dÔE;á¾°‰•ÒoØÙ<3Z£¸ƒ¥8‡ÎÓ”@ û0 p D^µ!=FÕbØØ´}aƒ('Ç‘r‡š€~ñ vßàE2¡/À&c‡pVõ7š¨©þ³0 cŠý €¦HŽ¨¨fwU0ï˜*ïðŽ¨ññ(°@/9-ï“ò2-®ðúR“¡ ²`2i©Â\@Y;Y—ÀFf“?¹”!“ôÀ %°òUDÙÅB7FEYC@”aõ”!“ ?ÅÒV9 ög•Öà I‚–L”pÉ•NÙ”J—ty—v9—zy•rÙ—x¹—o ˜É—FI˜‚Y˜~i˜‹)—êˆuˆ ™u™y9™û` ð’'Ù Psx|qЙà<ûðqš1 ¬ÈGñ@€ûDË )D_4À¨!ëþ·6)0JnFW>é§9 Án¡ÁöXr¡.haÄG‘Õi”fi‰_ÀÉß ž¤ø"Yšy\쨚¬¨žëÉŠpû@ (î¹y7øyO¶[¡Òà~FaŒ£ñ%  h¹óFÇy^Ä)l§+#ƒr¡?ÀcB|há˜fžxÕƒàI‡!ÚwXèx•è9` 븕py1*£˜÷“–7Ÿ–‹×g®àö‹÷Z(¬D%Gz¤%`‚  û° væ„ä1±þÒˆ°‚9buAUÂ:`ûÙ©þßIŠÉ‘¤èƒ¿á|0›ç‹°¡¹$FÐXA>–6ÖM¡B¡!MÎÁ q£WÜ TᎰðkZábZæ€uf: %*¢á)ªoÆGöù}*F ?€Ü÷sÞ…‚b $5¨€/zœU¶¹Ï ‘±©œ:‘e ªM‚ŠËJ¢=ˆ‡Bñ—¦}CBáNðç1ª :P«F¢Íu°Ð~í”›[Î@ Q;lÆÚ‰šly_ÐÜY¯÷J¯ùúõú¬A£3:£¸¨ªÑ§úÉ‘¡AÛÚ­ü! 6°þ•ÁkSº¾–øi©lqoØè®ÑšV!×8fà žøš²úÊæ ”°;û–a¯AF¡Œa$™`´¶NC@‚¡YØZ€¥ýPeì2õ±HŸŠiòʲûzµöj¢ Ó§ýÐœgb±AAŽ ¬"h¦` [D0¨ñNUú@©•:‘Ö³€ QÛ‰R°·Ëfµõ:ªûŠ¯üzpG?oñ…b{QAA:JöÁ 8®ѨæÓt YY³èúø ƒ|«þбU«‚K¯…;¸ÜÙ¯±‘£;Çz±º›CиAþ± 1 ¡tô ¶;0*Áó²tšôºQ ¡± q þÔ®‚r§Ö9½G!0-g²‹#¢ª‹µ"úºª»æzÛ@_¾¸ ¡«¹h«Þ 8P«ý#G¹Ûn·RK¡ Ñ´q ¿Ñ ¦¤û“µAÌeòÚ ¼º œÀ ìÀß4;ù+å k9¥BA@Ë­‹Jà` 7`:à›¿Hô[‚›´H5ozôgT+À[ˆwÀ©ëÀ«ÛÀúÚÃ]«£F1?«Bñ DkôÐU6PÂ;Å™Å%|;À ¾ /¡ÂróNÏÂfô3þxÐ¥µ5œu£¯©›¯=|¯üÆ@ŒB\Í°‹•zAÁk[,à° Œ°€»A¾]òus34ÆöD¾p‘hœu/§r¹ Ç<¼Ã ÌI¾ ¡X.‹ÅK-©‚F°ÁQÄA`»À&gÁJö§+(!k¶·ð¥%»Dn Ç…ëÀ €ÁE%Á!O!T=š»[ãz(Il¼æzüJûÒ¨®FAÃliþ·u”œÀYðÍ¿Î ,ÇgAÇ×·´3–­L©Âˆ¢ JLìÅÁ(hˆ0a±®O»†Úli´`ÀÉŽâ<οœå|ç\þî`@áN¬|`‚ÊJ"ÈÏüæâtž ʲÓ¡;ð ÐcV¬“ÜËàÜàŒÒ ¬Ò)Ò mü—¤pñr+\¨­"G ßÅ£>p¿Ñ Ø6rò6Öe×Û{h\½^šO[—½QÉâœ+ýË0-Ðsü€}¡ÑC-©0HP´µ D@EK`GËÎ~º*Øý ŽÌ?h¦@!Ò‰çÍX Ç-íÒ] ÕQÁÐøûÊF1 1òÖ!®DÀŸ¡Ñ}D€ ©œÑ€úÖÆñéli Ù<ÒRÁ{>2²2èÍ,íÀ~ýÀ‡{P‰ˆM_‹ý%F@Fþ gm°€jýØår+o­_ÝÂ:²,ŸÚòa$<ƒòªÒW ÝÐýÍÒЬ}ùðÉQ31ÌË‘hC@H  °@‚õ ß ©pÊj­ |Lš¥Ÿ±+õÉ®­{Ü4e$’Ì×ZAÝ)MÝÏ-àƒ­PÌBáÐRq­ PÏV!°  F€.ÞãÝÛBQ,û ±GvÜ€P†cȪ$¸¼l,‡âA ,ƒSWÏ-ã.ÝW-ØÖ}Ï€ÏB©q!Öà çááÕ@a© Æ-žºßVÁrW±3mHÉ-=àÝŽª nËuQþ2ÁÝ > lG°·9žÛ-(LÝä H"å.à^ãUŽåB‘/ÐåQ³A ?~(ózaÞÑpØ!â@ÁæQAoÞºà3Íâ,§ujü{ÎYÐéwÝŸ^Ý\; µ¡h®A.ž€ÓU²GßQä@±¿ºÒÿûÏ>òÔÉÆr»¾3Œ'ƒú àéÐ-ê ~åÏ3¹„Øê!Ô 1bæë#ûí«þà áÕϾRÇPÆQÁHJ’×-Çrånb@²¦½D þéïnì£N000¬^ÍÙ §Ñ?ÀêÝöÐ NÆï­N5{þÁ€†RA´,ék¾àðú'¯Ç.êð^ãy.ÂA ÷­îpêg1¢p+€Ú ÙÜ@ò€ú Û~ôçèáÃ÷ð—f÷ÀÍœèÜW}ìî^삵±íhð1B_žÐ Ônû° ­àk·ÂôþFýw#àŽ¥}ó™P@ÚØóÈnöBq—ÒÄÝì!ÛàkôE_¡Ð Û€ò 1Ü õ%¯Øf¾— ;>µþ\;ƒÃ±±â˜†é kaàâã­…•_ì•ùŽùžŽ. .WB ˆ^ÙWaÚ ÒÞ5·¡àú£Ð £þàú¢P=,+? Ý`õÁ.Üo†þ0î_ ·0пDšoùšùÎßéhßââYŽ€õg±/ä%/r¨ôEæ¢pògÑ ÐÑÑ/ÁOóVAÅoüöñþÀÞ†÷ZýùÏù>ïéÒOâùöäëWÐàA„ >kDPáCˆÃ…ãÖª•+‹]…›Ñ£Ág>tvhd¿cÇNF$÷„ÜJ˜1eΤYóä½'÷lîä¹RW•“°`ÉB´h–£D‡-: gO¢ëóÔ ;Eì¬n=H a3‘'û8ü˜²'Η\½reÛ6áZ…hÝÎ¥ifAP¥þE•æeš×)Ý•QCĕߣGüÛ|gŸ#H+ÑZ LåY(³¸6^Üy'g…P|y&ýPÝ'\a½Ôõk×JG” a gl£e-Ó#h0÷ [,ØV)atöfÞü ”ZÎg»9–@õØÙ_ÿ•^°vBÄ /{D¶{ÂföÁ, ó0®¾œÜ:_Ÿî=(´ìw&ÓÏúH¡´‹¿Þyà¦9¨€®¶¾o ¶ØQ¤™ý âg‘ed˽˸šŠå24±&¹Nt‹^üÃkÀì®pjJ(ég*@ÁilsK0…ÐÙc®fyƾgqG¦ö@t‹(Âþ0Ç&Uœ­Ê‚öqÉJ·ÖÀ§Ÿ‚b ;,Èt­µ2YûëÎî2¨’‚¨Éq. JƺŠ<Ò9=iúü‘Ìš‹) H‹KD"ljC}J4 úï#"ˆÀ‰¡ÐÔ‹L.*p"Ùúy`=QbË}T¨ÊèzF‘eô™­Ošˆ1)&Aé2GŠ(h)ÑQg!Ø­ÖPGR1ÇTvYfÿJ€³Ô 2Px«ó!V‹æ‘E¢YŒŸe.TŒ¦vª’)×¹Ö»‡(òû•ØîÞÉ ^«ðAàI=:€Y~™ µÆç,82¢bðü¶ÈG¢w«mÑ !G꣙ŽÉe¶uþ ã–w)¦w.Zù)<àø¤úuYÍ‚ÞIÁ* € €®ƒ„4}jävžÉHid™%m$bˆ¡)6œó‡–BŸÖÑzúäžÌ¡BŠÁžI‹ƒòhµ—ûå®»k#"æçÙØ«G–a§hØy&Ü¥'~ªÜšŽÁÃ>rh Š'¤xb )hñÅ—)ͦéž»¶Ü&Ó¾®m~ß–.n„)dNŸh¢F`]fF–D‘ ™‰FÖ¶öÀ˜&ù¼_#Œ'´f<Ê0"Ÿ|ò²GžÅ Í7·É „Ô†h¥®°Â5+°Û~¨ïEwŽôˆ„Túþ˜IƦÞ7÷å–ÉŸeøâ×zøÉ+WÑâõC§rÎs#Áø^ó½¥|OFöéÙHÐññü¯ ¨‚H1ÌeZ‰‚6ñ.n1<(DÁqÃ^wn)4j‚Œ‰±&z/ÄÂöhñ5‡|ÉÇÖÇB*l'nâLîñ>ûPrús‹?jÑ.åÑ&ø0CB¬÷$P{ ¬!·—Cæìð#‡¸•ƒdºÜËŒ;!‡/æ׸Æ}Ì*”‰¡HŸ5b0DøR¹x&È>büˆÏ'Eó=¥?yäJçW¨)Ô/yÓð|1<úµK ³ˆ#ib/…\Q!þ$S)¹g&,„ê< ŒÉs#½ŠD ‡¤É"­K+ù~˳äòšw\2Ò ¿P ?‚=+$S™Ë¬!3qHªî°2&ù #-‰õ5ƒ„g+»@6=ùoæÔ ePœiÃdvq™8¤ªhÂgÜá•,RŸ²†oæˆx f9 ø=e6é´!{CH˜ä£‡°æÈŠF«ØRŸ5>ÊQ&d (A=ÊÎú 4&ÎØC1ÐGF·H”¢+…?1jÎt t35èl¤¹“CôÁžŽÂR>ü`R·dPEdé&JN,ªf¦M˜)MÛÙŸúawðJÆæÙU«ˆþÒE‹^J@¦2u©MéyDZ“vSE”BË8êlU®'§G2Š&S¬z%«šPÓÒœÕ&iÄNMDŒátæ{œkc¥3L»šs¯~UfÂÊײFÓnIk€š¡‡ ‚4ãR0WjZÞC‰ÖG PÙeN–²z,in Õbüá†eNnMζ”ccV.E‹+Á50V!­õÈdÛuŠµ¶žìVœÏ6t.í8Ä>;|:V¼ÌQÇj#r׃äÕ²~]¯e/{Ùév¦ºl± gÙ:—|XWíJÇû_À”áXA¯A à^÷²—½vo|3_¨»ê»þoOÐXÏÖG¨æp[ŽAÄ‘¸ ÎepXÜ^Ë:0·íM;Š‘Ó=¢ü•Ì‹ÿc#„¨sAê‰v¼Òÿ¯ _‰1õ5[·—²íU1] Lt¼»}Ð-‘ gd9˶:DŒc\ Þj˜È&³U~ц•ˆ¸Hr‰QLÙ@st›åR;´LFbdËaVQÊÊügžàÿçUM˜àfÚ½Lh2è hžü¢Ÿ¦4Lð \BȲŠFô¢SÌÀGŸ¶4í;j'jj³•Ã¼#aîÚ­èCÃÚÓþQ¨;›Uïç¸úÜ5ØÐ0à•´"64ƒcëbšÖ©ª4þWÂÛlh?$Ã…‰šÙ¼édwúÓ!µu´a2ho7»¼3Q3‰9mO#xÙlyr¸‚&º»Ùdvµ Í|úØúÆw¿ jǹ-¼8ÇPDa€X&F^›¿ùÍodÇzÝ\i·¼ @‹7[Ð5Q3Žýq}CßßJÅ3Þ\Ÿ¼Ãh(ÇYæo˜Ç<ßÿžÍ;„˜<ãàVùxwAí™(àå3/ö¾‰NsÒ¼ÀùAX|êÞà¡å&Bu>«î(UïÄM1€ÌGÞï€ ª¤îÍ›nL®+M0ò@uë“í–ç˜iÂpˆ0@ì^9ÉcBX‚é9cþwÖ›”Kgsßêá7׆ãì$¾]ç;ßÿ%(!ž‰àw6xƒìÃR†gì~žJ0^«Ž7Û™{_½WÞëŽb‰¥ódx@A@oD€ô½ÙŤë“úá›ÍgxJ| {¯‡ŠM#NHïSpæ÷)@@€`ã€øÇoþó“ýåW?ùÙŸþöÃ_þëG?ýã_þú»?ÿãÿýßÿÿÿ³?ù Àñ+Pñ¿ý›?ýC@„?@¼¿úK@…¨¦‹C9dºÞ¼9œCÕKC=ÜÑÙKáÈ;>,DCL3@=?<;.ö>@=@B?15ùDFCFHE;;÷IKH0AùOM0KMKNNFNOMDAö;GøPROQPSURLHõUVTCLöWXVINÿQMòZ\YKRö]_\YSò\ac_a^SXõac`ceb^YñefdfheX^óil^ikhd_ð_dólnkjnqnpmieïkfñdjòrtqtvsolðvxuipñxzwsqî{|y|~{vtñnwñ~€}~xxnƒ€{ì„…‚|~톈……뉋ˆ…ŽŠˆì‘Ž’‘‰’…Œî”—“•’Žê”–“í—™––‘é™›˜‘•ëœž›š˜èœ¡£œšëŸ¡ž ¢Ÿ¡£ œžç£¥¢¥§¤¥¢æ§©¦¨«›ª«–¢¦éª¬©«­ª©©æ­¯¬««è±°¨«ªÿ¯±®±¬å¬°æ³µ²µ·´´³ä·¹¶¶µæ»¶â¹»¸³·û»½º¸»å¿Àª½¿¼¿Á¾¿¾âÂÄÁÂÁåÆÂáÄÆÃÂÆãÇÉÆÅÈåÉËÈÉÉàÊνËËâÍÏËÍÍäÒÎàÏÑÎÏÓâÒÔÐÓÕÒÔÕßÕ×ÔÖ×áÛØÝØÚÖÝÚßÚÜÙÞÛàÞÝÔÜÞÛßÝáÚßâÝßÜÞàÝßáÞàâßáäàåäÛãåáææÜäæãåçäèçÞæèåéèßçéæèêçîêÛéëèêìéêîÝëïÞëîêïðÙïïåíïëðñÚîðíñòâòóÜðòïóôÝòôð÷öÙôöóùøÚöøôûùÜ÷ùöÿûØúýØùûøúüùûýúÿÿèüÿûþÿü!þCreated with GIMP,õšþÿ H° Áƒ*\È°¡Ã‡#JœH±¢Å‹3jÜȱ£Ç CŠI²¤É“(Sª\ɲ¥Ë—0cÊœI³¦Í›8sêÜɳ§ÏŸ@ƒ J´¨Ñ£H“*]Ê´©Ó§P£JJµªÕ«X³Ö,†ª«×¯`ÊK¶¬Ù³hÑžzu*­Û·pãÊ}Ûj®Ý»xóŠ­ûµU+}õ L¸°áÈ+^̘08o5gœE¹²å˘3kÞ̹³çÏž[ xº´éÓ¨S›N¢Hµë×°cc>1™2* –®AÛÍ»·ïßÀƒ N¼¸qß™EÖú°óŒs¨=ÇØ£à pÔ²kßν»÷ïàÃþ‹O¾{§I˧/t®¾bôö«¤V%[ùûøóëÏ>=üƒìýÑ{F$Ÿ@§H2Í~ 6èà}ýÑ4C @!DoLsáC:ÔA8†(bxÒT³àxÎ4á†]ÂbCÅÄó"C² $E'ÕŒ¨ãŽÔLSÉ.ä¥(ÓŠ3i$F°Ãã’!N“,A¢'á‘TV QÙ@Ãä– : %ŠRªø"?ÿiå™ eã…–\¶‰Ÿ—Qú·¡hÄG5u ”Ä?µ¼áæŸåÁ ¦œš™ç¡ ω€6šÝ4è˜#©9}ì2©9èHã1‰è§/0Ž£¤Rs̨¦ª*ªâp'Í5þ¢„9$¨´nxÀ>Œ–úç4ë ã+:•þŠÎ:šj70¨ŽB(…'jH6‡^`HŽºV+hwÒˆƒ É.[ë·ÌQJµä^»Ý«4¤&Þ‚ë®U^èB®µOºªÍ:E¤ÊE,´ûî¿Q±`ß¼¥šKM5tp‹ª3Òp “§Örh5õä)®܈-Ú]ÃÊ#©ŠO4Ô8üÄZ˜çËX¬dƤ¾ª4°¨ÊÉ1ÅšìÊ÷ÜÔ×ä ³£Ó0ÃŒªÊ›Î-ñìóÓHýÀæÐ1K“BªmÈ#4Ó,9- ¡PÇ BSݦ4ØØ£ð\£´v\¯ä5|*׉F3hã§þÙ€¾jFª)äsÍwq«4w{²ž ŒŒgBQ ß^ƒÏ©ñÍÛÜžÒáawÞà” ¹ŽÒ¤¢ª8Òˆwâ{îúP¨Ô7:“Ó˜ƒIª(C²xš£Äùsd‚ýzV ,:;Ñ`s5ª¬ç”/Bp¨xZ99Ç“~M46c^ï'ý>ðÃ[5‡.ªgŸŸ4ëp¡ªi$ò ~à›$~ùø³ÄÇ+ê‡È¾úê‘÷¾Çº“åï€5©Ö¿ý ­>@U ŒàŽò®€;›‘ …h‰L9\÷#.àU"P6,xA!ð…(QTúFèÔQª’ =úp%þP ñs NÐ!UfÐÆîT¿’Üï‡PÔˆ^6ÄñH#ªJFÐÔD’<1+º8T6îA%ÄcjUüαj&Á]ìBlڌꆦ1´ìH0a¿ sU‹0‡Žº8’/Fñ‘E f8DidKU€`$âØ5DZò#³øÃkh0¤ ôXáŽ)CZ%x—L /ø°IîLÃ;H•¨ñÆý2$¦´ Ït·#Aà­ÌÚL‡* ˜£µ´%%å6#FŠŒÓ?€VJ8,¦0’ÉÄe.•àÄFeMq¨êΤƒn ’\†óâ¥ñ^¨êzþä;?âΩ òC™â<íƒ?Ò!¢tÓ>=ÒÏ©0àPxÎŒÂP uš­DÂXÞ˜0¿F-´# õ'@…‚y5riHÕÚ@,G}”#!iKÚHqü­˜Ú˜7GôÒÄT¦>$Gò™½i|ƒÀ@/̱ÓAzss3‚Á¡áÁBýã#ìc,S• ¢ºô©¾ªX"€wX”\ÒH$µÑTõT#?ë Ìì±/ ©šPG°·f$®rÅ_‹*t¡ªÂ̘_1ئãPÞ #‹8P³–ªŸÀ:.7´Å^¤±LÙ¥•ìø"xct}ìåP‰^´µMžµþhë9Óò-uâXÛܹØVd¶´…Z&2aY7IÀˆ ª¸`Ùù6!Á.pà"åŸÁ¥ ($qQyLp´p›sÁª‘à¼èUج‹ÑVI Ð`‘#21´i`c¸=^ûÕ“á¼’0„!`pÞ0„½G1Ä¡tñ} š‹F4¸ *Zð5{Ï-Èy+&}üã¼ApvßuVKSë°™2tÚ¿ Dˆ<â‘¢b ^mÚº×ÃÒ¡4tñ@fqŽ ÑQÄDÁn_"Craƒ'DU®!Þ ¹ ÷Ç9®PàÏH•hʨè(M #³©’Gþ•ƒLÞL½@Ä5ƒ1„áÎWÈ9P13 D!‡*úŒD dƒˆŸbh„£„>ˆ‚H+!“þ8^iá b Þàô¥òix£FAB1lÌb ªˆò‡V¿:ÕÁB®!1¨!×ñÃ@x=_ D׶@ª‹b;Ùói¶@Ø0;{>þµbm; dÙvÃ@æ€h„{Üi/"nÈáªÃiIrÐBUpHÚ¯LJÀYÈ Ø1Ž‚‡c û˜G Ò`$|áiø?.r0á ¯8BØpÿÃâáx‡þ=rƒˆü€¸É’r‡œå$ÈÉÿqò–| ê@ˆÍKžs™Ã\å=HÍc. Óüç)rÑ 2t=ãœéHÿ‡Òÿat‚Ø|êUg¹d¥îô+T4fòÀÀÚ„a¢Mòû <8oÀÿä%ÊëeƆ34š…ýóìqƒ@ãÕàPŒ€ …àÞ·u<Á°>ˆÄ…[yåÀx¸ ¨@úîvKRà :¦p×Ò`.­Í€$aZHžò•?d509w´¡˜YXÇ=eÑ[ÈLd*ÀŒÙžzDòãXø“4LñÇ (cÍž/èCÞl£×8ïCÍÔö¡ãPàØúšþ¼%´Ý;UÔ˜}ò•’Àù¼ð8ï_ä^*‘¶@ ç–®Q‰T5âKãßÞÏMÔ÷þ‡ñÇ%Ò ©ò ÓÀbù·:(A>Q}ÿ犊€7IÁúP\÷¡ h6}d ˜9ËWHÒÄ€=s @ (x P@]HÑòÇ#Ò°¦P¨€=´!íw$ðЕõ(ÞÀwð‚#ò*øHXExÇÊV'¸ q A&I0 ¡ãJ¸°'$ÈÀ0&6"Õ2¨’‹Wƒ6(&è 0oßá “GŽ K(Øð]` ˆ†ú‚¥´†= :aÞ Ó þD! žƒßÅ” |¸)~ˆK#ø4ÈpÞ IºpCqÅ°#Ò€x8¿@-‘؇.Ô|—'Å÷påAq2c 2 w*ÈtŠæ1‰íT‰=#ü`QÓP,µH:ßAp uhv¾ÈO€˜nÀ?ùa‰ö*`‹ú! a‡BgÈ‹Ÿ—ŠÓÈúàŸ˜ži=q1ÏÈÚ \%`wâ‚ä8’'’ xÁ÷ÐÞ è ~ÛtÒ…5!P"" ¿À-³Ä_´ UŽ6Deƒ€à1 ñƒ6‘$ ù ©ò¦X¨xƒ §pn¸)HàþMm:¡ä€F ’-‹€*_Ðy&y’j˜’1‘ç :/ÓP yÆG€“]²º‘Éׄ ñ„@qGu2g”à ªs”ây;±Kà”û1a¨’ R9• Rst(ïgUÀ à–áQñ€0 r "Ñ Ô*?)‰÷ø[‰&]ˆ ¤)ÕИÕbëà˜9â=áf:¡7þ“b¬òŽlY˜B)ÂÈ(×@>Pš¥9¦YšiàUäà‘1± l"â¨â5˜½è™Y¡d´¢Ât ؜؉ È—Õð G 2ŽR Ü2Õ°–ãG• a•?þáš3²ƒqÙ0Cvà€::2 ‡€*t€›à ¡>Á‡R 3'P° â1ž‰ €†" " îfÖym SŸéù¹Ÿ“(B 8á%¹åðGéäž„“ >µ ,± h0©‘äá t¿ä ÓÐ Ü’Ø ¡ª›XÁ›‡ÂáÀzáq ˜ÀÜ‘ Í¢—+‘¢ r ž4Ÿp ùŸ!Ÿ=±Šubb&ñÄ™®4ܹ€=]ÂFEà0š›(Y 4z(`„"â X`i3q>u( ö*‰À£î©¤Õå¡*Áй#Øœ4±? Ò Ê^Hš¤þ Wvš ¥$Be¦à Éù͇ oð8 2›I¥ _úž…úW3r hB üø s §ÚL€š`ÀTäñ×'÷Àr¦À-{@¥5H§ x¨'a§€”ô´*°Ñp€âá¯÷гªF‰€*y€«¨«Á¤)‰¸Òªòª²¿ –@Ÿ2¡z“æ”Tð0¨ H­^Æ«$Á!ôÒ@Å°Ð ¶P¨ ,4 ãðP2QVK(ËÓ º©Æ¢ ¤¦_y¢× K`‹Ú°U€zt p*Â@ž(0uL&Où1 ÊPL·©°ÛQ ¸ þ™ £Xq¥,ò~pªc ´Õy.åM CjtNÊdd–Q€*´À®É Þv° ”¢}g²•¾é*hF’Ø2 ÀžáqÒ<`–äÁFLЬƒì Q;+ð*ÅÀÊE„*OPAßa ©¢o…¨ Èæóª׋h+ŽÓgð£ÿÀ¶0{Jeò) T6Þ¨0á¸×@€ôتÀj/‘Œú‘®Å©´­Ä ±á¸+ 2»!l_=@¯f¤oT C^J·<0©)¡H8 ΀*œ¥°ÓP ª ŒÁºR{!‹&³ÀÞ°ßñ þhE×."p ßq œ¤!Q ~Ðëð>@îÉ‘lð¼Á¼nû¶Á x@¹§2[DEz¤Ý;‹+¡ ¬T²Ù”¦[E+jº¼mÛ)ð À¥Ý¡™Y•‹*ÚðìðºÑ${h‰˜ ­i4 Þ€ Í’î«À,2¦Uòté*ùP†‡{ h&è{ ’ð-‘-Y';"PÜ@Óp°h&œÀ­ó"C{&s @ˆÙñ †õ±$r€VÜ ©bVL·ù¸°Ãã1 Ä€*>9˜ â a»'œÄV¾/¢PL »@1 â¡ pÇy|DIµpÐþø§W@4Œøá¦Y¬¡ÓÀaPõ¤kl@ Ü(§Ú ð3Õ` œÌÉö´\¹` ÇàJ~Ðk)¡LØÐcŸ£Þ0»ãà‘œA“ÌܸèÃëšÝ°F€*®åŒà")' ¨å ¿ŒÎó“Óp ªðÙس,G/B‡ò W~Š*å Ý«pX)«äÆ'1_‡8;`p¸ù§~à QÍ•TËa–ºæ 0ß I©cÅÓÎòÅeÇxj`R÷! ¥°\ ‰ R X0öÌLøœúì*߸/:âP@0Ò@PP©Šþ*¡€Ë†ú@QùAºöÄ“”K°'Ä’œÑ±ÑÛ0¨b Éô*TFe³)Õà­4°ÞqÁ(¡MÑÚJý*Õ(ÂõÀ¾ûýM/2gyâÙŠ¹µ› Î9ÅLJàI9'ñÙÐýŒ*ð¼I rK,] Uì—'ñ@D•Q »‹h= °À-•0@¸Ðn%áð&ŸÀ-O`Õùw ÷ðåƱ×aõ"‡&ópj ‡õÑàl%$ÕM@… Y”÷1 ;@Óõ嬞ýaµ "ÈÓÁ?P’ÒðÉpP×ÚÑ )PÁ߀‡;êGþ@µ ·qÌ@9;ðͧHŠkÑÛp^9Pj!<Öª*\$7ÀÐÙ©¥Ë Õ`Ö´Á$Ê›X# gE2¨’圀ÐP p ÑÛÖèuúu 1 g»Ú×u2X3cÅ¡B3¨‚ es»M¶u‰*@‰àîkv {ç¥ý—çàç'@…é­8\ád† ë9Ý€¾ ¢·0¨­Þ0–0Ó€‡‹ÈÀ…=~ÛUð+h¦w^ ΤÕp^= Ù‡^ï*Üñ–ü Ë£©ã‘‰ä1 ù2ãJàPÂ!±]g¥ þ¿<ÛTƒÅxõ°X®ÓÑ ç74ùbÐA&°{9^y0`bÙB›‡>"ü — 5’t )EçÞ3 –@Ï!a‹xÉ@›@Åù'—Ð!œ]φÞ<°5~^ˆhd! §'“Ž8}&•¾Ó0ÙÃ*:"|ð—À0`³âᨂN ˆ j ñ`øÁ¯d`Ùi¤:Ð üXè¡VŠ^º—¬èelŒ.<‘'°ëÇ"y}$2`bÓ@¨ÒªÃ»Äö!ž©RÌ€9Þ0"¡˜³])TN Å å½- ŒðèP€v°mŽ\Þ2ŒUrþ Œ+Òª)PÆ@¸ë÷?pì>Ñ®•PnÞ1l š ÁÞJÓ Šàò# µn¶'䀕ÑàÖ »÷DX/zfR äúǸ‚Ñó…ÇØǸßFö±)®' µ&ǘmÿþ æ]÷IÀÁ§ÿ° ™V÷GÀ÷á÷€?„U0Ø,ÈÐhÿøD°øñQ ÿù”oõ0–/ÅðåšïùoÅÀóÿPúÿ0òq H?ª/UÑ †ÀÜ“T‚ÉçÐ * ðÍØá1¼SóÒ‘¡ Äþü/'ú³P7ºð²ýÿp¢ YðÝ TÞ‘¢ožÙ?§P7u1Ó¯ý¤P7³ YØß2€ö‘{È€ëàô¬¯J†ÁïŸÀ DˆPàÁ 6tÈð¡Dˆ VÄ8Q£Å‹9vLH1äFŽ"?fDøÊ5–Ô¤U›0Áž´–5‹ ÉcYMžÔ¦-šPˆ²žä>Ôóx’$ÂyÈõ„êr† ´EÅšUëV®P§Q›EAéؤ'‹I"›Vm±-jåp†ZºuíÞÅ›ï^¾}ý"”¢«æ·)1ÅA 'ö¤%P_£NÛsQžÓšIùëÊ4Ç\Åþ4E³ëhÒ¥{N«§ÃOfº¸Ð²NëVñù°mß®«÷nÞu=x«‰Nòáž“<ÎkµjâRČƜyç-Ùlc)6º˜˜œL¿uZ8½Oº6Ÿ0ž  Ø¡üW;ý|úõí³þ]VÌ^¢kVsƒ—±ÎpäšžhÁÁ$+BAh8¦%C ,HË%¦"ˆ ¯CÙñëî+½ùøÉÆ­÷>’D_„Ñ> Ž¥k”ˆ –ÓîÉ ¶z:k l´ÉpÖǦaGØXÈ®«hŽ‰ ‡«<ÄÒ4hŠ)`–MœÏ­±ZŒ±LŽ¼13Mµ¸§³kžˆ‰–ÊšTþk)€ä©ÉÖñ¯%Ô$`í€qF.&zúÌ2Ѭ)Æ3Á4ï·Ð ´RJå@ˆL5ÕtrSO9:€š~y. lz’dºª€š¨°ù"¦Fª*›-–ù+9²éJ|bZãJE‡= rš8#MH{ëÀ-gŸm/ÓO=}gZk µ¦UbšÂ@ž P§®Ô‰JZbRâÔÓª©á¯ ÈÁ«ovˆéD‰%6›$ ØtYÞ Ð!`&BÚkFÆlj§%iö›àZkZæŠxêú¡¬`‡¡ºf‹WúR•x±"¦„ D  _b§ñÆøa{ü¬ @Yx  3N:ëÖÑ•jtâ ahBR› /À«4Ø€CLö`NV&sàC\@:K¥H@GìÍKgÊ‘vŽµ/ÉéL?¿$uð‹5€ceŠÄ§VîtwÝI8K#þ•M Ad%“àûí± Rø.Ì#V³6Ö6É P&`ž¨{~À>yBN#éÔŽÓèEú‘F!N#~h×ý [Æ:Öµ~é@\YG7?]6“pDT˜' ¤B£ÓðáH»¾i܃º€âÚVKŽp4×¹Ï%‡K_K—pL×.–è„c&aô©:° 0ð€•¨N€Å—2b’s —qŽñéW LyºEÒµ.Y.›_¥dÂ-¹F#b’ žd£H:ª(á’»‹Æ7bªž“I[Â5¬¸Z“ÖW&•.“Å¿A€y2 <Üæá8™9ææÞþD™ƒ*“¥±ÖnU­ sõ#&É !Ú—Ö–ØÈñÃ)ZÒ"TåÍ3Ä·ÆöC:W—¤¢* d¤>ÜŠ>ÌâdÚXО¼;¹þÅ„®Æ˜xÏÕÃÞ÷¾7ÜÒÒYgÛ߀¨X"á,×$ãÞÎld!&˜Xh¢ª1×hX8F$‡=,€*öûÖ˜ùwÀ[BðFP¾æÍö/iÐC2ù´8åqãckK8ÅXWëU³¦ÈŽE8¶Eî‚cÃ?®D9oаÓô“žUÔ4Êûˆákx†Y Ëæ4ÓúØ>‚7"(ã˜TX Óà:_–0‹¨(ã9"è…Ë”h Ôñ³äÀÃj´þ꺤=-?ÿ·c¥ÀírܸÂr(Fz–†Ç”·øªx—Y™&Lrïu4 õÝt=ð‘æÇàAgã |þ¢zlH¼yšq†“M0}ÂÖv1sI"®%s< Ït0Šþ®‚¿—áQFr‚*»® 4z?Ÿ€#‚yºñ:*ž¨kø0 ~ï>‚®s¾sÄYR¸‡å¨kßÚÙ^}MÒ kž@#–ó9¹NsÀªh9KßH÷/h߸shúµ«Í%ò»Rp Uñ¸Ïs´lƒ–¨†¨ò×¹G`´0©¨šè†¡™€[¨:Ûµ 06;‚†+¨¿“¯î{¯"…‡ˆ‡Ê‘5"CÀ[Ëa#ªØa¡†Ó¡Z¸O“o›SÉlH † ôhð†÷Ã=t¯+‡ˆµò£5Ã6þ–†˜ØUò ¤¸)žÐ&ˆ“ÅQ†N šH† ò2­¹†Žû?&L?€C{ˆP]ÂHø#³Â–È H½©Epƒûð‚iê‰jˆ„˜È‚w+£Gh¥P@HÜpèì1º‡€'ÿ;ÀH«®l€+”†5œH(‘„Æ€ Ù먣†J@>­¸3x¸ýûŽiz²®ÕâOp Ð](g¹\(F]À…jÀ À{Á{R€•c‰‹šÚb‰l(¹€+ì‰hªcÀZ´î™€,-v@þZ-ظxJ–»€Æ™úºH{—–(B *»qþÐÀQ¦ 8„r¬År¡z™Vˆ8«‹‡ÐCw”C„È€°È‹ÄÈŒ,€C¤GëÃËGŸh7aà RðÅû8¦ÖsŽzsÌ mÀÀk´#) ÂüÚ:ú¨Ç™òl[(¨¡á®š ¢hs‘+P²W)¯ Èq`JqÀNhJ¦ìi°eú†éKµkQ’ÈOÉÉh´"C0–èMš’l p`Î# IÃkèÀ  Ëº¬Ë Ði =#ìsð€²êÊ´¦'L@j±PKëB„£¤f:( qp!‡-x>¨¸ɨ„u(‡rà„OøÌÏl/FR/-”Eþ—©^è&#û b‚+Ø`Ò+°ÍÛÔ‚ÝkÁäÁ6,À6G…–0£6J}E‘jÈL‚˜ 0ÅqIžà›9ƒÒÄ S@š#û VˆÇgÙͽøJ|úHþʽiÄ Xƒš‡û*“l)Î ¬‰N‡!» p0]4i@­rM&,†#H(&H‚5ÐÝÏs´{T³ÂPÓ?c‰q€ÃQ¤42É„O8HiøaŒ ؃š‹:vG[- è£ÅZÌS^0 ˜ ài¸83)"Cᬕ1…rˆ†r0‡|ˆþ!ƒT0: B È‚@Èþ¹°jQ:jÀ†˜ˆ×Ù‚Q,“Xš4ì'ɸ†“Œ°N ª† µicÑlKPZºh¸( e‰L“3x®¨Sp¤•Œ€K\Ÿbð¶ÔÎ'e-ã1G›šoÐmMÅHÔnƒØ1“Sp„áêKjP³|¤cP†iÀÏ]Ô;sRµ ¯úªPð¬Òüúj(ˆé‰ °U[u.Ø7 it VtÀ>5E¸µùÚ¸W}ÆXµ® K¹  HµÔg ‚Mñ€zXΪ" vؾDí &Ÿó7®<2^Àƒ%骞à…+ðP¿WFšƒÔþkSEm¢ `OÅT³ñ|-ñbÀ9Š×–h`‡MƒKÈ×ï¨lsÓ‡`DØCGCGC†7`@e[X–×O£‰%i‡<²XU4k#XhÅ'÷|L)½šÐ0YÑCY®\V—刀Y²ˆÌ‚ÏbbÐüˆˆ+[ˆ‰À¦lkyƒêY­ðø·ÕZ¢°ÛKp ½‹ *£ t4÷ø SÝÌTôlð)9ø¨MÉÚ®˜g¨!—™†Sˆ6emYVÝ8K îD[¹`=˜? ÊÌ8-?PàÏ¿˜…Åõ~ý IX\>P׿ðƒÀƒ7xƒs€ªþ˜ÀFjàÌ …‡‹s°ÜK€ ]°Ük€€Ü ²¿9Ñ}7|= vƒb?˜ÀÌPÜÙI^Y\ëXä\:Ü„XÚû¸€i¸7^€yÐDŽÐF)d˜'ùÞˆð†7À7/8N¾pT¸·e¯ý‹*È€x(\PUh‚v˜›šhEµ¸·NHÁ¿87À·0hÓ¨Žr€ ®EØ_Ux˜oÞ´D~ø™{£9¿já€]wÚ½8FhéÜøàœN€4X‘½€‚ö-HZ»‡ørˆ ˆfXÌM‹†zv°µ¾„‘ þÇ(„M¨ f$J0Ì“¨Š€‚6Ú }À‰‡8øÖžˆ`¨X†à‡Ú)®¨‚†ÜŠe¨˜Ò‹x~Û‹{ ²ƒˆ10ℹؑ uD@Ip‹¾^µø¥Jªå %Ž×´à‡zØÉ„¸d~)¼˜ŽLâ3h‚%84Pƒ¤R“0³u\ðͽ8°ÒY`¤t˜€$ÍÆ #‹+È:‹ÐƒFJˆÈ‡ØD ­2.0ÆÓjåüŽeƒ-8‹@³ó-ˆ$èå‡0XOqSÀH¼8ä´¥¼º+p€ÄT gE\ ,¼€C6²º¸\ŽþxÐ&Ǹ+oè‘19‰mþˆ*ª‡#Œ€É¨ pˆ¯9¨[Pù±S® ðâ]7¦d®‰Vpƒ¬¿”Õ<.fS‡X»`ÙÈçU }žxRç5©^ƒ1lÈ$¸vXNhx=p„‘ è“ÐJØ @ÙzÆ‹{œ‡(†LЂ`<`¸ä`‡Væˆ$°¼ºh†÷ëÚù†0h™ix‡% v.DÈfºý"€±€ª„D°ëDhÄG¸ëD0jP rˆ¿Ò(.¸.¶µè“h¶Í€a,ƉVŠpvˆàl ˆlþ„¸€?(ˆ-øâeØkhÜ í8ƒØŒkœlè=ð€•þdXj á‚¸¢¼‚ø†¨™1ómý‡*ˆ¯bX¸½‚ ã–Ѓl`n„X†wþ‡è>¬Ödn~ЋeŸð‚K‡boÄY´–€w†ú­…¨A¢.ˆöþ¨úÕ…9.K4„h`ƒwà… ¨uˆ¨ƒƒNŦFúý‡&ð5]Øb8¢@ †Æ¾f¸ÖPl€9ÖÙЇ$ØF\E 8PïjH…OñOð¶ŸqZ "À† Áé®ð­æoØÆ –…ð¿þêH­øÜ~…š”~ò‚Âe/)ˆ”îŠn,×òÀíèò‚ør„`€Vs„àÀU•Îæl xn²€šý0j™FäÞ `‡l¨ºj¸‡¤sˆÔÿÆ‹=·¤¾‹Oˆ{¨‡ ˆêª2¯Š¶\XaŽ`‡’¶ 7h#X8FBœÐqj8‡INì“„>6ÚðpŽh‚€3n‰¿Æ / EÈ9<@ì‡Pl¥" ×™Ç\‰^¯$>gÜp)À¿DB½ ˆ,†Æµ‹W „p »‹}ƒè)%ÇÓz…¾Ã Þ 'AKg¤Cþ˜€p…%„8…èúˆ>ä‡0`Ý»r‹àÈv°ÿwÀƒ8âÿ (@0bˆxÿþ“¢ÃÁ„ *äW¬A‹}ì&²30ÁÔ5j«AÇðã¿j7¼$Ìv$Ê…ºPtJhÉ®Þx€ÓhÓæº ænjœN]ë5aôŠ'қЮI¹§Ô…7’ÔKiuß) Gœ´þêõ+X”¸$… àžÓó( ¸öŸY g¼µj)Ó´¤z÷R›OCvfÍ*¨‹’Ô`< +œ§gÖ5(‰ßdËË×fµ&á@~H\Œñ ç %0a‚¸›¼ºÔÇÏ+®Ä,@3t“ØÂÙôJÃ&Mï4v(w²|Ù&¸Â $ÞG{abÀϧö‡Yü(ŒçmÜ…t©ÔãÆÛñô¡‘Ëaõ5ù„ÒiŸ"òçšOuúÕUVß7^§Äg•"ndÄŽ~ýœ†šj¬ýŠGJ×;Æù7Í8û©“‘zÓè’Ä?ð5ô#2$¢‰)ZÖ Ñ™J]áÍß­àç_zþÓœQMŠ)ñH[1nTÓŽÀƒBÐÅc!:§arÓ4Ȕ䄲àQ$–é²X|€¼ò?úËï[^Ò_ÃÙf£2,ôš ËÜXsOÅLi"çlÊÊiJ{Ž:‡å*@ó(ˆN[-Y+à Ö)}Ù8(DȲú¼Û!5¿ñU8&^‚WR•œFÚÔø¡ËÕ^y¢ÈÏœNSÍ5¸X2_Þ…'ä2±[{uš~½×–[€8ª„c„u‰èqê !Ç‘Ò„„¯-¡KRâ,ˆÎMÈbxJ”@áîžÓ£Î=?üà†"¬ë>âÃ*ž’ªù8äÓ\snŠ„§4Ã7ªþ $#•W5*Hx@M¥ŸnÓ4†€ªPº»ÿÓLpf“M<˜¥Ï<´Ù+¾½ãú;H˜KO¼^ÐÔsD< ¨¼¢ TÈYлR¼ArÄçààU¬LqtDÃ&ÙèüBŽ+\ƒfÀñ†7¦ñC¸28Sá'?JÑ!^¸Dñ—#oœ‚a8Á ¶30ÕbŽ3`ôât OçÀØ‹8Dp e£)d8N…ì!¥ìx< ‚Z´OLá #ÕB…€‚g“a¶®Žs¡ÌÑÇØzôŽh[L7d —ç\ ¼IâQ7jcU_\È+Œ!ÙþdÀ…¨uÈC†Rc$ˆÜ@E4É'ç8G=’…3„áŒAƒU°3^&å,ÅÑ' #–â(Ç ¹·å…å5ËHBŦ· Là‡âå@TaÊ*" U¸4¯°…0èa‡IÊÝ$U”5]r ( 'E=^ƒ0LÛPRÒqÊÀ<áùÎx35zÑÅ@s"8ozy8 n"¯ÄÃ9_ô8 Ž=~‰ÙÔ&Ë.y#P‚á솖±…ï‘g†pV4ÐaŽ‘šãQXIͱŽ\òä ШË>z Žê%\˜@ Pg]`ð¡>ýéB*™¨Knñþ¢@kÅ Â@ž&P"† »£páÄ°4á¼àM7СÕnœ&öÐ*:Ú1 ’µ¬-Û&¾Z¨ƒ•%S†[/Ô d~DZ áÃ+âŠ%¨vH0ÂJÉ=ìP‹¾!e :,b°¦…èì`j¬cŸ#T;ÑÏëÔ+¼Äo4ìúèÞRÉAàv "Žü‚‡j‚D|èA L«ihc¤òDWíAR_$øXd{ôà‚f²`úPÁN£R#‰œ-5¤ÀмTI˜Ý‹4^)ŽºñõB×xEFôÌ2ýäK“ºú\jÌb¸Ä}ïîŒû%Å©âJ̵‰sæ—À²þ`É;,Š%i\ãÁ°Å‚z X°b``$–¦1 4à Å(5P€f„³Ó8Æi¤!‚çVÃ…/Šó&_/mí ~ )só«_pƒ A~ ‹ëòDò€CR€Øe!ë¨`–®Žqt o@ƒ:üˆ%løà4Âq¹—â-ëlÅT‚™#@q^£Êøqç>r.>@O"ðã8±˜"K©Â÷ÕK™Ñ˜! âe“íMvaTO ú‚6ø" sì`A>¨F¿÷Ó``'ÜÂcN)…›ˆ.¼HÂ:üp¤D²ÐÛª£Ê4XPVzi¬# 2‚;R^3Ÿ&!·‰–!ó­Ç­÷¢K0Üm õ|çÇ!zÚ´¡68O@Ò˜þ†Ôgk,(æˆ{f•8zÛd^D‰ÔÅu°zðgeÙPP±­ìf¿ mµ!ym# ÝÐ=& ãíbAùÀûM¢±½?ïPºèÕÃtb–€d:ª®ác/!šOè:<Ï7‹{ˆ§ ¾÷½¾cpôãÄœ 0ŽÝdö¼‰2àagM#Nä4 )û퓇öºòÞ›ÔU—Ÿ¼0,¯ yhž Cø±ËþÓhƒQÊxÂôeØ TÂÎÚç>—×­ øížî Ú5I>_(ÝžtÃœ†à Ý_þqŠ2”œÐÂMxƒd‚U´‚: Hþ8Âöy߀à® Ž$ <ÂiôÊqŠ4ôÂa9 5X ¼Hƒ2¨×1yÅ;ø *D  Ƈ ² îE5°B$LA)”,”‚-05äB”Üiƒeè ¶äÄi|žjÌÆžaXaÅßà5TÂ!&,H% ”B¯äBˆÃMM@jÕ7LÁ:€Õ:\ HC>Ä °ÙÄ5ôF"Ö°L¶¡%f\L6p¡NÀ NC7p¢œM@­ÀI7à€•ÅÊ |å ¡AÅžRG%^"-vƒ4|B¾]’9Ë'^Ã?Š< "J: ¢zôB L@ OCþxyE&¤“$þÃœàË,Ò¢%Jƒ;0ã\9Œc9Œã:LáipÒà®M€ )`Ó4ÖcP]#±d£6²à-ÀÂ/Ôš‚-,p[‘˜Ž¤ÀBj€MÀ8þ9 6°<`ß) ‚=fäáàã°èã>ºá!ЦÁäœ$œ68Ýi¤B)$¥[f¢FPC5ü£2ÒÍžôÂñ}B.ƒ–ýd=&mæa–ß5ÄÀEBc¶eÀ¼¥dj„9ŸŽ\ÌLA7h 9|&hJ¢h‚i–æEUò™'La>ŒãÝ9Ë-¬[dN&5DÃ7páˆÃ.ª>,>°T'¨Aojäo2Fpîžùuƒyž§yÞ¥³´Ûaß<“UòE4$ƒe¼fO`ƒÇ½SEâˆ8pa/èEù„ÅWÖcXr]xÆxŠ_5|&¬Â(š‚&ª'œLÃQN@>hÕ'|B¬áàž¼!&|(&þ4B€(&TÂ:4$ HèIÃ!°œ2$4xB|§=h]$(s…T)äÙ(žÆ|f¢6(&pH%ô_g5Â#,é#0é#4NñÛq„”i€À%E1 ÖèXR Ž†ÓÛ­\{28`ƒšŠƒ¸ç(\áÏ48è0N© Ù'ä­Ãñi@7À䘩2>” ƒ—~é4Úè[Œ)14äA*”TÝHÃ7|ÞF 8ƒV4Ääi4„á)þèi處úp¨MHƒ3$ãÁ~&|XÌ‚CÕ#%X#%Æg£®*žFì~Йñô‚ÜäÛþ ƒ„Šœ9§- žMê^\Ð]«ÐሩB^5èÝd«úØñù@Ú袼;a$J£á)*‹øª¿]ƒbꛕG5äjžÆHkÀH¾½“3ðD$ i¹¸qÂB€r¡_l+äÑC ¸'-¤5`C.¬‚-ß¾åš_Т’ »‚£>ŽÍÊ;ÝZêY*¥% >Àë^DƒéX£ìÂ.Ç<Ü@L(„!„ ܃[ťȉ«-ˆD›>A§&Å5œ~l¢†i¤Œ¬§b,ÈrX9„6+÷¤WR‚^^\¼"CdzœQS.Ý4$kœÅ@Ò&E6 ‰Ó>þm¯ºáäÛ4\g‘Dƒ3xk%¼í€1CCú@nªÇ5(èHxÃàA¶öÊ5àCÑFåFA#¬Ãv]†µ†%°k$¾¯®ŒÔ6 ÕFœrŠ‚-H"PО|ø–«£@6x5èí`åî5ÌàqP‚­Î-†ìWŒ.¼,å‚°êžÜâa½h„ƒ0hj½æHxnJ8%dŠ“%´ðF¢ðzñ‚64dDÂK7°-ôB9è­|~¤^mz@8ÞW(B‚Ùeƒ·t¯÷Bí„PÜ”ÍB„o¯¸ÓiЀé­ç4(ÿb€ oRpãaf™ cÜÐf8Ecþ è4ÚØ`úo|\ƒ‹h]Bp£`ƒ:î@ÂË0]¸œ†DBæò†6(ƒÞùêq\Ã)hXÌ*À]Ã%(çRï‚nƒ ?ÄC·.Ôv€àý É4¨ãP`Í4hÚtÃ'@*Èë"E>A¾T’_}MÇ4€Ï^Ô™ñ¯oŠ0u˜$D˜Å »ìzNeI~M"|B«Êƒ<-È4„·_Æ$pït0öúÜ9H’ŸaÈú1|¼†'D¯$:ÐÁiÜ.M!‡^9<òiÐA´A¸§A$CÎ=ì1u¼†HWö¦Ç% Ò&s2?Gþ ÌËB\‚xAŒ2ŽTƒ,H ×Ì*C\qÊ™-ïâÈ8¸Å€Ä;´À&=Ž0P_q$‚0,3mx€˜ B ‚ò›Qƒ¼Ó.píeT3RX*:p à€AtàÃØ$’/àA&D[Í4Ãäê0“ /¸3hÀ³<'=Û³3§‡4ÐCª.!2v]zš§ÚíI1\Á@ Í44í? LA$ÄÐ4%Û4\©™¶¬Æ@hA,Œ D¡A¤Ù@ )DR'LD1RÄH5ATuŽauB<ã?ƒŸ „öUÍ@€µXwõ@˜õM0¨þA°sTÄ¡C+Ä5AÌõ@@AØ5^„^ãÌ /p”^³s„.¬`ÿõ?$ö@èÂ,äµc«Âßýƒ.(®Yk*ü/´À „öÐ4•pôB|t3ãs–X*¤ÞòjÃ)l;Âöé>/ qè”3¼è‚0_4 ê‚F3†iÏs=§öEMÃ/(Q XgÃDC耋œ€]­x@ËƒÞ ƒ‘‡00£œÞCw{I1¤K£\C´Hù”GÊ\Àð&ŸCVæk 'Ú˜ˆƒI V¦Mƒ74´£“¸èº··Á1Á4k:—z]áÊøMòãu©sß÷ZE¥ÛD2,È!øóþN0ÐùºÀ@£]È£P‡K";·+•÷H¯[j$èk¦»úôÔá'Ê,¼_Æ5èÇÄùŒN£±C{Ý~Í5ìŸæ »fÃw¯ 0ÐŽøv§çúÖízJT:g0¸»^ÃèŒ$ûq4¾ ¾£D¥ƒ U¢ºNƒ\6¾ð?T¬Ä‰€'<.h2ë`Ž÷.¨¹‰€C¬£¡:h´§È(KCµŸ1p½BTÁ+hpñ’öëLèB+$?.´Þ?:æƒÔ:”ØBÍ@C< Hð|ÃdB²íøøó{ý½ ðê±\*¯x£„´Béç…^—?ýw_í› ËkWÉÎ/ºÝ;@üX Þ4j&T¸!µi÷”8‘bE‹1fÔ¸þ‘clj’¼y9’dI“ÏœT¹’eG\’ZÆxãZC› £)Ã0G4i7ÚÄå†_Çf(Ø ºô ·%¼dF•:•jU«W±fùR«HšLmJ[7ìl`Ñ‚$Ò›‡h:Õ•n]»wñæÕËU¯Å¯pŠD6ͺŸ€oNÛq$ª-Þ/|¨ oeË—1g¾Ë7óßÈÒ"‘ ìpä†à:˜PÐôÁiì(k–MQOµÙ·-Ám™3fÏ€¥iIik†Àt”,šcYqÀÓ®5ÑWt·fÕ«§Äž·÷åßh¥ÉBÖ”¶ÒƺAvò ÖÏï°½~}û÷Yv·ülþ5%¢ÅA¯¡iÞ(f¥ CË eFÀÏÁ!ŒP ý*ão)il!KÒ”,œäT**‰lœKŒ< ¸FBîÞYÑÅ)¢°/ o§ ¤)±Cjι@Ÿ˜¸GGÉÎAƒ1ëR$íÒI™dÔ‹F›¦aæÆÜ1¡iâ9r¥¢T %›jt„.\$€‰º'ÙlÓM•¢Ìkʆ”I¡¬oÎÓ²7,‰JG`ðãhªfŸ`ãÍEmt£8ñšS¡“!+<µT蚪òC€s¦ ušWøÑÑSQuÒ»$MèšTDË'ÏL=¡äªˆÈ•ˆ-*B$Õ¨tiñטHÖ%˜4þkõ E«¤šLš¦•µôÆØ–¼¹çZ–šÙV£UíjUšk„Žš Mˆ ¼m×ÝwcD¶3t2E´>Ê™UËiôaÞܺ$•æÑ™F_-Ç à‡!^t`ºæ O´4²L¡× °ŒÓˆ5b$$3 #≻šS0D“gáLýð„ä™i~e­hü)Ñ8ÔxË0T¬Yè¡w»9+ ¥A‡ÑVñy¡bª°¬¨5‰®Zh£±²ðš6x¦æå}©iâ2v­þ‡ù¬þády}£wÙn-—¯3eÇËËþÇ‘‘ó†뫾Ç ÑhqÛi‡ø¹»ïŧêo«þÊQ™ÁþÈb†îL³i°ñÍ9‡’mïÐ}U´ VÒip4¿Ì¼ƒíÞÇ«úJš_F'æšË3í„‘ÌZɼ“X[6š®¡¬zÁ=Ói,鳩]þWØ©º!XÉdÃmʦ‰l¤_üŒ¨Ÿê†hDDaîOCaü÷áŸðs˸X塽F74s$o`â‘7TO3N Ë#Ìþ†œ£w}ÑEÞ²¡­²-ƒ€™ ‚Ú`8nL ŒßW>©Ø@ Ü C®Ðu"ŒŠ²¸'…s°†!œ_…4xƒ,ã Ââ]Þ~W6T°m:d/´£™Ž•mÔÈ[ˆ:#.dü“þÍ­ÖÄ'Fq?9<¡ÝnƒÅªi±lPôÛ gäE>äaÌ‘7µA hÔ À‘ =î‘}ôãÈlƒ,耙HE.’‘tä#ÙCB’’•´ä%1™IM6òØä'AJQf£4å)QÉÈ´¢†­tå+aKYÎ’–µ´å-q™K]î’—½ôå/La“MªÀ)ˆ¦‹N,óe»‚‰¹­€•B£FM´ Nw¡AÚ¤fÍj‘M"H¤ ÙD&Ét1{BSžŽè=iö q"I€þfæ©^ @(Éô$Q zê1´oRÔ]õ4'ÈÂ1RŠdÓ@BS)I·eRšã¥a©H ÓFÉtf4íÝM]šS.Š§$ó©E€Z³šBúà0 U©B¥"EÙQ+’Tš-•©òÆAj«F «+@K•*Ô®V`hU/zÕl~L"MÐê̸ºV&QM c˜@=@èƒ%Z^õú&¿B¬Í? MÁ5±©Z,ÄTÖÔV­•}Q6(4?d@› øƒÕëÙ'Õ£û Ú=X[ Vµªµímq›[Ýî–·½õ­Gøú[áV "bîqþ‡VÔ^e³@îsñ³NÈ /Ë…îuße]ìn÷ZÚåîwër8L h€@FШ€"ŽÈ€Ð 8O"~àÀ(†¤äè@6;@­œB- %80Gð.Ø$Ê=R6ßqÐZHä8Áj-(ô±+D¡1Ð9xD ¸‚DŒ_¯x$¶©’ðIH¢ñH…³9ãgÓWÿà°‡ EP"›—ÐpG²¼8T,v²G\,lr`"eű­$âìÓÇèðD><ïNä«éÁFŸœæDùÙÃDÈ¡Ò +€›±D6ûc0öÅ%Q¨$þ@_5º"l®iœTa÷Qdš¸èò—ëÊg)wv"è é‰únˆ¦°=‘F?ÈísIà)rhšÕÿà4E<}c°—Ñp´žù)é6SZ"º°g«5ýj8˹È9C6'Œë#›¦ž4IÚ‘à À&´°%ëàøÍA0G…ž ¤ D³ÅÌëÜ™¡Ù‡µÕŒmhÇÈAx ÍùÛÂ÷Î7º_¬,£!«ÃH$M PÃ;Íòþ½À‚3XÔÆáÀ4³©Œ¼z±ì |s´"™õÃD*œ:†3Xaèö?Â`²‰Ìc I¬ð R5¨áÈÿ(Ê8Π†—WäAoæ?.æ‹c *œÈ,Ô à•O]ÑUQúÔ±^’ªSE2Ïú×=²u©¸AÓûÙ9rs´¯/­¸ÂˆÙw¹Ë& ;gpml-matlab-v3.1-2010-09-27/doc/f5.gif000644 000765 000024 00000067466 11422240325 016644 0ustar00carlstaff000000 000000 GIF89aã‹÷þ )))85,,583338882821,0 d>>DDDWWWSMMfffkilwwvXccZYY–¶ 55Öÿ÷ÿ6ù66ÿVÿvýr-¢EE«UU¶SS¼vv³ffSNôEEøUUæMM÷vvÿwwø}}÷ffäffÆ-bÿ|¡qÿ–ÿ¶þ’6¿ÿ?ÿÖùúÛû)ÿ—DÿšJÿQîŠNÿ Vÿ«Eÿªfÿ³wÿ¹~ÿ®qí”m°ÿP‘ÿq–ÿxšÿzþp¬ûs üaþôDúõJþõTýöZèôSþöfÿ÷vù÷yèõpÔðc–œZ—¶..©DD®UUµJJ±ff½de‰Öÿö6ÿ67ÿ37ÐVÿvÿ9JýwwÄDPÿIUþVYÿCKþVaÿfpÿhrÿwwÿqxýchùWGÎV3èe¯?ÿ¿pÿvÿ®Vÿ¨—ÿ¶ÿ0œýw€ÿ|…ülöC®ÿK¨ðS´ÿY·þI±þf½ÿk°òSãÖÿùú)ýØmþÐvÄÿwÇÿtÐüQÿô[ÿõMýðbÿõrÿöoýéWèÛf—“•••µµ´™œ§Ê““ú’’ÿ½ˆÿª’Ú³³þ»»û²±ù¬«Æ¾Ÿœÿ„Žü’¥ÿ‹¦ÿ‡«ÿ•µý–ºÿ¦¬ù±“ï®ÿÆ™ÿÉŽþø‰ÿù˜ìó’ÿÙºýÍ­Æÿ¶Êý®þúªÿû»÷ê±Íèš™™Ó‰‰Í««Ú‹“ÿ˜™ÿˆ‹ý™ ÿŸ®ÿºÿ¼¼ã«±ÿ»»ÿ±¶ý¦©øˆ¬Ñ‰Ìÿ™ÓÿŽÍü»Âÿ©Úÿ³Ôú‚þ÷”þø‰ùí»ãÿ¡ÿù²üø îÑÌÌÌÝÝÝÑÑÚÈÆÑýËËýÝÝìÕÒÔÿÇÕýÎÿãÌÿüÌÿìÝÿüÝäÿÙÌÌéÀÀåÌÐÿËÙÿÝÝùÊËýÌéÿÄçÿÝáÿÂþûÐÿüÝñÿÌÿâîîîèàã÷îîþîîéÿãÿöîÿþîòÿìîîùàéðîðÿîøÿâþüøÿöðÿþÿÿÿDZÞ!ù,ã‹þý H° Áƒ*\È°¡Ã‡#JœH±¢Å‹3jÜȱ£Ç CŠI²¤É“(Sª\ɲ¥Ë—0cÊœI³¦Í›8sêÜɳ§ÏŸ@ƒ J´¨Ñ£H“*]Ê´©Ó§P£JJµªÕ«X³jÝʵ«×¯`Êõ÷N€±hÓªmúN€³kãÊË3‚Y¸tóêÝ«r€Koù LX#?Àü.̸1WjP#HÍÁïôU¤ÀÀÅŽC‹–úàÝ»¼ó÷À?ŠïX­ïèÛHû”Wðµ^Ïþ€ ôÌÏ’p‰\zx€Ð ˜y:Ö«OÇNýïÒ»þ[çþÝ;øìÞ·[OAûíß·'Ÿ ýuë ⣿ÿ¼xõÓ±gÝ{öÉ7`vúQ7ßzö‘§|ÔPà~ÓÈŸxþ•wáw j`yéX!‚Ö7`d68(`£•P€Tá@TP ˜ @zÞ5ð¡‰Gva†Šx]F. AÐÑ`©À–°å—^€åVgæ@PóŽÇ=ÄOXBPsþ Ð&W„u‰œ`Ep8YÜãÏ9køƒÎþHâL)Hˆâ òt“ƒ@ôœOA™dâR&!hJi6'¨ÀÛ&w*D¨Z€êª¬¦ºêþw<8` joŸÙæ„•X—\"¬A€®vÏùèH?ÎH‚JGÔÒ„?òäÀ?(`SP¤€ª­Æo»¬™ì@–¼s%|Fä×[ü—@ÀK¬±` ; ‚.{Ž>ƒ ÃN$¨ aÄ6LØãOݤ ‘ÐÉJÚˆ@®?üdƒ‚&¯mÒÃ7­ÜëЩòÖüê@Iö™?–à€ÂÇæÐÉÀæd¬áúÂ_USž1,PÃ8+±#©Ä6NxÓO«ä Ñ#‚Jñ” B<—–{nºë‚ãÄe:¤€ÍxÓ‹þHÖÓ.Ð=Y˜#P¾øC4ú<² E …+þ¬„@òtÀ[A%Œ|R<$ˆàù@Ÿt ‰?ô±Ã7P0a 1@D3Þò °ß~‚E›îaÑ–OâÄGâ‹ ã#îÜxÊ£þtÓÃ@.xbP'% Äéè©,ª?›ä‹7KœRı?D€ª¢Ïúé·¿¾ªzŸDg¹ãfÿK„ï/|øƒÌ"þ˜„4ŒP„RxãZôÈ=<¦‚ƒ¬Í$0ÙnTptuCa‹ÂHq#Q(‚º™™úFHBõ½Ïñ3Éüêw¿¦$ñEü'ˆDE£F8‚=–ðštÌþØ ›A2A‚’ÄC%hÛ@CÈŒŒë æ’^åvÐz4ÁŸF@‚)šŽA,#lxƒ@zú… 5¨*P‰š’@t+€K`µ‚‹3Äh ‡?Šç,’"‡tG,°Ä$„‰G&0p²±éÊÁåhaLš*³Ç„>”‘XäC 2‹þ¦ð:Û»Ö}|E þª•Ý}Å·XÁÂaø‰rÆ),Ž&x#´íÌÜæ ¢ \bt:H‹©ë '÷mÐ^÷À†a|vƒf€|Í—W¨|ЗuftÞ”­… ÈGóV¨Pó&dE-ð`{ÑòÐoÃg'è K0 ÿ9 û@„°Zhð 1qaÆŒ§QÉ—„è€Vˆ÷“ †%hpþp È% ‘0ŒB üÐväAv¤“o"¡ 0p”vàЭp ZsÓ ì þÐmÐ0[bQ„ ¡Í·„ Æ|¤öPþPQÂ`‰ãþÀ°ZS üÐr¼V–"ý0{[•:?ð üÐ M0 Œ2tÍ°X‹ ÷À‰ã€¢?8fÎWˆWqjŸ$«¦‹x8‰ƒ?ÖÐŒbÔ@ P@) Dµ$Ú3]«à=þ{­ÐJÈ4 ‡° ý i |¹°æ.>õ€æxŽæ˜|çØ„0ñ„‰½˜‡¿˜JíÆý7à3'xhïä+ƒé²døhápƒ§ §˜Z„W‚(¹@€ˆ@ˆ¨ŽêˆŽÏ‡ˆïØBñ(`&§ËiÛPN°k6jlÁD¶ôD9`p?Gþ ‘fÓ°Lí€^þÐ l |K7Ž‘1‘‰ŽiŽ¶è„ Z0½Ð?&‡‡Š"dF0 $†±0Kåb† @qA(NÀ GPE¦ð ’ÐÈ„§}px[ðÁ€Ñ´Žx‘i;1!}–WyIc:&Cqr–g‘ãPP Oó “‚‰hÀF?ð >´ð:xÒ@mþ  k`bÇv uC9”Jø|!ô»5}®˜9C3¤8Ëà H@^­@ dp¡U!™ ·'¶ „¨0 ‘Ðv|°‡þp€SHšå˜ŽxyšGþÙI©”z!˜²)œG`‘`ð& ü€Œñ™A¿ä «ùRÃä˜IEÈ´ ŽÐ™¿§àq€2„”!ˆC¹§ù€ÛÉÝésgäÐF‹u ( þ€CF€×¢h~ö›[uw7pÁ :ðCPP–g‰§ ‘ÀËxjÐ^JÇtA a ˆ æˆ <º£ ª‘ÚrE§cç`í刨"Y³l4pçWw÷¡¼ægÅD Ëe1Û Ñ ‹ŠpÅ ƒ³Þt£>µ?ê£ šÀ—0‹_±‹[a§ Q¤2¡h€Önú iµ þþà\Ñ]œÁ‰በÄOQÀ ¥x Ò€‡R›Ágç ppÚ£çȦnš‘NÈ[Cjz |*(Ððcûx“˜–È=dX(p16`I«@Hæ4X  W¢À Ò¶XªåpP€‘ 8š£ú£ŸŠ‘° Ñ §ú©êGZQí%€šx\˜˜f×­À…7R î°Ÿþã–þ° i@³Êf*a»Fˆ»lÜÆnìÆôbûԠ4·¸ÃV§Y¡ÇÀ"Á´þò(‚`<µÖB¦œö¹V+K±‹cƒ0ØT¬ªó«ŽÖõÐJ7" ¬û TxŽï €$±m·ûƪ Çhbvœt[H‰Ç<Œ¼*gþ‹;žýPÄæ„ þPµV¼Œ Ó•°@(9K@ óð:äYæ˜þ@bÂW Z@—+Ѿ °ÊàŒ»ô’3Hs—À&µÜI˵¬ë+À?ì¬(`<½ÌóðO¤ †z¨–üžÂdo–ÉZ# ²à ŽÐú@dÚo cÎúÎ)áÍá ÎD›ÆÛÎO1X‹¶}%G®¸ö°4–ÃüþE?ÄÌÎ oŒ«ë¨„‡ä Ä0]Ѫ,§/¸^¡Î\!Ô3!͈ñ?±“+G¯‘N&û˜AÔ a{L\ûÇ“ºÒðäb1Ë4 (ŠÄÛÀÓu[¿F¸¯i=¾ñÑŽµ”;½µàÀëùÏQ} ÁlA `¾rt óÐ ÍÐ7I¡÷€½0=)jˆñ«»—­»m}­ÑJËp@g–\þp‰lµ±Ô¡ñ]aVêÔa©»`ÜpP«SÂ'ÛLí›ÖºÛÛ½m¿¥êÙYË„ÆKí L-„*ÌV¬ïJ²T«1þ5oÏl1¤0,Ú \r&_‹2±Û¾=Þ»ÙÑÙÂ]É2¼IÍÝÎpŠÈY² P@ ~-ßðõ)®íȹÉíàÈ}(xhkqJ—Â5!ÞäMÞæ]tªjºX¸ kï-Ñ"àUº»9œ½¶xš:/I­0oüХΠíP¾š! k@³ïm6qÖP¿™m°ò[° à°é 랣ý¼¥¢±×J+P@¶K ~e³Ð€ÜíZ|4:qàýà>ÅÀfnæg~æiΦjäÁ=äQñŠ!³q(ºÑá!hþ^hŸ˜\dõä± ;°Ì¥ÛàÀVEÜÐÕìà?óMi šþÐ6šÞ\°hî梺ékÎNè-ç=ÁÑ°è npç©d›*µ’öÌ=Žé‚þΣÚbë†þ0Ïv¨°Š]r Cb=8\ õ`Ê3áÍ›®æŸ¾æœnæ¡>@ÝT×~s=Å}(šqg(C¹s¹¡ß´Î€C™QT¢(Yšl²‰ä‹à®Å^&»hÌ•àæm®æÍþìÓ.­É;q-±GmäÞî²ÝM}Ýö]½&;ⵞ«¡Ò1¶þP¯ÐfÙ}…8õMh0 ¡þ¾Ê^û~߬æ ð>îò£Nê81›:Æ©îËðŽàðši¹ŽâK´ Ð+°pôJ/ºÑhèƉèþ Bï‘Àå@¡9iQÓÙ‡@ÁìiŽæ_ÏæÍ.íBNóê}ó¾ ¡‡ãµµÓr„S|]S*oÑøß‘{”cž]<øúèÕ\ï‘­©7ÛõeöŠ¿øŒó1óf_ÔñÜí«þªº¡Àqß±ñ z÷”Ò=†~÷ÀOP°k²pƒ²Ð1½X=(ò”þ^>Ù)_ ŠÏæbÿõŠïøQátá+Aܳ¹ðíã˜àŠŒ›ÀÄwŸ:`ƒ3þA Ô°vd*·±ópgä{“Æ3ÆZ_£D¡ã ! €ûbû¸¯ûù‘ÿÄó ÏQLõ2%ˆéZ멳û:SsÈaD¿fôù[ö¿{lzùƒˆ/‹.ˆ-^ĘQãFŽ-"ðh±’%MšÜ°ƒÊ!]¾Ô!Lš5mÞÄ™SçNž=5ŠËr¯"6Ñý9¨ïÑ:ÛŠAUÑ -Ž:VU\áÖ°4ÖbB˜½SHŠ” ×/’3ˆ‹AT††\E]YÌùÄûäË*O®ükR@^œ2 FœXñbÆ …ئQ¤þú9b×´’¨£¸þò¸Š†M9vlꦱž)ô†U¤*~ë0ûkhD>lòUär¦ñâ½.”ô€\9‡äÌ9´ÞñÁÌèÕiV³žý%öè+¶áEùàåÌNgW¤åå*¯i-±ÅÏŸ*$HŒpó')ÒAdÚ1d׊̙H;€i¤æNZp9è ´ˆê ¤Ð¢w*Ä°¢ ‡Ã%¨ïÂóç¨ñn §>RäóÇ›%|ª'Š'‚!†QêEa a$}yËaÐ8‡.»2ä A¾Lj®ALîÁ #2J)§Ä+., P<Ën›ç#N„¨ž%êéé=Z†æ#ê3F™F(Affþ€þ"6øø!-„C‘ÊšŒ4nIå’k¤æœ¤J?e´QŒpÑ¢6 1¤ajž5ÁôGLoxj 2ù9eMS˜!&I êï uРԟj t”¦âB:.0$DBEk6Ø q¹ò»]´ô‡ËMû„‚ªÞ«¢yfDb–z€Dvìq4Ôò.a]4¤–D7]uÕ,C&—0jâMl^Å‚“”R!’D[~H±ožŠZie§zšˆ‚Ld©ïˆTüqbB¤¢>ÚÈÓŸ3„£7$`R@PBEÙPæÚÅð׎WfY'.pÉYIØ⧔€+ªE=ÞûÔ{Hþ‘”wô‰ä™aˆ9ÆR†Ò8")j™£[=R0]æH^·WUŽºk¯7ÚfˆòYCßÊú™Ù~LA¢!ò¦ …oáŠd1"oYüY’̈٣Gr€ ÷k¦î(dçÖôjtµÖŽkÃ'ú-Äö'5æ q™ú‰#ÞîtÌœZi¢g~P ?I$é§ó…üñÅ·ŠÎ(–ò‹çè\« uÝ«!Ïn:Ýmâîøë»' q*ÊG.ˆ ©LŸÛBG"œŠøY"Zº›(¢pŒm }"‰FÖþb÷Ç74¼\ù‹™quƒo’H å‡iCþ]ò_O€ò<ˆ ŽsÈ Äˆþ2ó¯ÍlÃ"MðžMPŽŠ âOÁO4ñ}¢GêHƒ0*2ÀÿU¤\TC×Ö¥B°ðqD’\ e˜!ql2þ G “ B@¤X )ÖäÀŠ446©[ø|&D$˜‚D[UçêD;5€ "¸°á ýQ€úÙÏ…-\’ CxiÑŒ‡n8¸ƒø>"DPa–½ L‰4™`E¶±&$ì˜b•Òüñ†6X_fäÝFÎ¥B22…ôñò‚ À’°ÄEbxFN2R7<‡ò„ A@Äz™ñÇ@Dˆ¸Bg49X¶7G$ˆB`Ñ€ÄüÁ#ˆ4$V–Ãœþ©‘E‚ј_L×$Ù“wÀ’p, ¯wu#ö²¦E°™bY¤vy:F·„JZ²Ò9³ Ï*2­5B>’hDr³¦sþ !'éÇ/†Œ.\&O&€\¢"—2 ‘MfÓ¡<±’EvQHˆâÙÇmüQ$˜³rƒ%dy·ØÁŽÐBþ Ë^س"¹P£5‡™‘|À¦Œ¼i#ÿiS”%Fš =(>ÆÐ2>Ô¨;ÑB.$JÑ]&Ô€L¥}Ì©¢¹¹¤Kp–Ú8Z {øC]„8ïц^  "¸sèL1RÌœ¶0§oU&c,™<ˆðÃ’ièQõêþ’—a©iþ2àHNò™SLUõH‹V‘p­u| ‘W)Ãhj^0;©Ö‹Ô´­8…«M#Yj$àÛt@jð¼T(3žQéêÐÙ¾˜’‘>!Xü1r”a—Ð3«vÏ" «%~Ô’¾íav?²â0rá<£îÓ%žÝ©hµF›>èïx2â€mî$˜ÐþŒ@‡²×%Í#`æØ[…@„ÈGVfQsú Ll @YšH>¶Ae«ÔƇ6Üð X¸¡COØ„ÖÂ9} ¢á‹`b>yÇyà¿`'Ææ;’—âk¨ÅQ1i],cþ¿Ø1†‹i|cãØ:ÎñŒƒ\ãûÈ?2’‰ yHŽÏ…aÜÊSþ³?®‘ä#šÐ‡^²’Él‰œx€ Ѐ›ºõŸ- Дq˜Ã9€{kÂwL ¡§µH^÷ºj‹ä3sÒk.:äóǃð# ü½ˆ+œàÅVu<~ô‘¶mýžäà€9¨A o€6´ÅÅIΊä2°° °ýVdxñ„ xb‰ôÚ,*«ÕÝjþAälà:Á>gD"vØëoÜŠ àÕ™¨«ûÈ–Qœ ‘eŒ@Ï3ëQ«]lÛçöµoú €—µy@ykòŽJ `¢®¦Q5nÍ‘kä“\ž¨™hLžˆ…Íp‹—¿\Tí?è)TuŽ¥W±@¶‘Á †]ä3 Ífs¨×…«ÃC‹íQã¨ö ÒS›4ß݈ª×mÔˆVäh'! Ó I`æ2wAÍ_þ Óu䪖‡,JJÄ×µ1=ÊG–^Ê…&ܨ ‡HÃ!nÓm3^ÛëbΛîTS~ìzÝ“ExÁÔpV„Øš XV*â)Dþ¤¹Å¦öC慞ѻ§?YÃ?ñ[¼p¶#Ÿ˜wXRÔ©–Àå×}¾…©‚¸(DÐ'¾5A`¢ïTê:Bð]„Tfqgo¡z]’à ÷lÔ,Rû´&HÛçGú®í€êä–„üï ‘_¶z­­Eš×åïÄŠ¼U ¶ðMÉ›·¾á²;´x˜"`3¯‚„6R0Úaƒr5‹0‡qЫGS¤Ç[?¼¶õ»¶öà ~°ùCÁŠˆ­‡ú=NjAø‚X1¥Ýò!Úш½féC¬_Kðˆ‘„6r‹ïà‹89hÁ3â@P¼ó ÁT?ö#þ#áÛ«q ®±YË*¶)ú#TºÁ/éÈ '8"¨›4¼›"8[¢O˜„xâ‡Üà-²‰‰,„ˆÛ‹B)D‰ó@®óËãC-‚”i‹‹x«/¡k£9*‚ì»(`CŒ¸*⪫²(‚$`JÀXL0„)/dˆâ;ÄÛ+¬B@´Â »C”Ÿ²ƒ‹5P¹Ê BöA…5錋@CŽðÁ‹˜›y,ˆX‡A°dÈðQªˆYá?ZdÅW´ÆXÄ{¨’ã¤ò⥪ˆÍ%Ar†(Z›òÁˆƒ©D‹PGŒØ#ûÀР8ô‡°ªeCº! n<£§3ôþ›ó HL?dŒY¤Åɹ­Špƒ$¬¨ñ‡µ¨+!*‚þê&ÀÄ‹P‘ŒL%ò!f€ƒ7„Aàƒ9 ƒ=0h5ˆÈ…3ˆ0j4¿kÈó3HÉ3Ä„ÔwÈ‚`Šø?Z¾l¾:bG¬ˆl¨W8½C‚ì{Hp~à‡“t*P‡.4$ŽÑÉ>4¿œ°”±È›T „ìÊ–qµ|0Ú´Œ8òn6Ò=¤Ä Ï)‹ø™ú¨£0DÆ4è¸ø–º²pLKxBš"ÈkH£ YAÚ¿¤[#Xk‡FLÍø’®ÒHéÈ“¬b,>ˆø*þ „3ÀÇÙÂå9ªÅ\+õ È°ËÜD¿(CQ/Ô++¹!b@È ]ú*ö‘… éMxĪˆÀHwô’ÑÉŠrÊzŸ8xŸÂéº L°Ïñ¼MÜO³L ´LÌq9ƒ=„ˆo*º8ƒµaóÀW‚Nˆ €Nä2…{臈¬·L>¢ø+ –˜LLV4OñËÏn»Âõ¤œo´8€ƒŠè<~1G€é£Œ)à‡(ˆèë¥ýB‚T z ܇?H‡±iK‹Ø ÝðtÐ5O}PôD õ´QF ?èùÉŠJ¾å’–¼³‹hNjȆXÐþè(…Ò!HÙ (eJ((Í„8È„Ô 1(CBÌ õ‡j,OuPóôÑÃÐF‡âÇÿÉ'j¸4ðB}06ŒÒ(Y8ûÈA‹ØÈŒ‡ohg¾p¸ %ý#tÈ êÑù¨Ò¨S-².sÙQ%ÏSÂR5•’ºX#—êœÕ>]b˜ÎÄZ`ˈM@JÑë7|gpùX |5`·ø"Õ5ÅQò|ÐÜlÓÕ¼ÕaÅ[ô3€ dÐE9»7ÖyÊæ°P‘O蓬0QÈÁhˆ"0ˆlƒµLg­µñ´wýTñ´eÅ‹°?v  þ­ˆ£ƒˆ õ‡fˆ§%"ŸYHG#âÖ%øOøV†)NPh:è%4hI)}%V¾Wy¥×xוW{õ‰ú;ªü‹Ž‹ ò£‰…dˆŸ¬†I>¢a®ñ1 ‹¼Ä„­‡—ƒ*¨Tˆ~@>e†à‡<=´Ê ”Õ§bYõاýX’í‰É̦&dŒî| –} qÀ‚øª,ʪ˜ÞòˆÔ›¹ˆaü âY9θŒ”: Yó„sXƒØ‹ÂÀZ-¢ÍΠ˨ýX©Ù ÕX˜ÐZ—àÚXWwû¾éiD ª5Zr¢‹XÃȆ!°!€R@`[ûÐmѨ}ñ‡þ¸¸¡§)Ü(VeÝy%Ä­ÉÉÔ½ˆÃ]Ùa° á G5hšV‘¶ƒˆŸ‘ ƒ½ˆ<Úˆº‘Ux¹|9"ÚLI…ˆÍ³À,@EÙeE‘×ë…ÚëÕ^ªå‰feWÚõˆÄÝâKS>HÂ÷9BËȹl¾þ“;ÊzX‚o°˜£æuöA £c*| ŠüÔ×êÕÞÍÞ~Wî݉9ͦLU ðÅ4à^(+î3ŽØ?è<8 Á~ †?Ý [ºuâÒ³ˆX¨Nx€ˆhxý'ˆ±]ŒŠæô‰A;ªM¥šx}ÚÞaëE`ð^ †hƒ61 â7PÙË¢` ¦à þþ‰Ø,Ó††ƒÀÿ8G#˜ÜÑs‚½Ü^'Ù8‚¬7 <6hš÷ÙÍ6à6àΉ Væñ½ˆ%¼,´c•AÈЀ-B°Y¸ˆœí/––@u¾.× d=½ÇÙScÕåØ.`í…Ze%ÜHæ:Þ;î%Hö7È“#„Öó‡xXΧTÏ•ÄBÀ|ÈÐ;‚XˆxšAÞz•{ê,h:MfcÖa`†cœ°ZkbÚÆàdðä þŽ÷1‡7ÈC€?0ƒDð8èU,†„uåž!˲Y®âæâÓÙ9‡Âì%-ƒê8f3ÂáŽ(`¦çíÕŸØþ•¡½m`š°ãµ !ʘg„ßø¹51Jý\ J,м,e Ë³@‘GÀ¹- â-‡dÖ[Œ¸-¡¾ ¸‘6à‘Ž×‘6éa¾ 9&U~àÎ<Ž‹ q#A‚§÷IåRØM=„¡áŒFÚS“‡•šEø48ˆ,àN#>bëðäãùe’6i’ÖÞ‘Vi›`i5…i >®ô‡^°n‘'”$1#…®=B J¼ˆiØ 09¥£î_ïà—^Œ¨ÖÛK€ë­j«ök{–E|Ödž0‡-˜Æ|h‡ÌÌ)ú*¢³ÙÐ5Ãke—pk‹Ø%@¹Ö¨ ô…þ­ þÿÁëŠèè¼ÐkÊéŠèk«®j”l¬&&¹¨éEtd„ð~±·…¹ #ÈÁyÙãz‰V©h3†úhóƒCðˆà!†tH—̈mˆ8mиë6¤špmïþn‘ön’ÎêšØêÂæ¥}ÏÍi‹Ú‰V¶)‚Q@ù€8Þp¥‹CU~@‡¡ ¤±)Ò+Êî¬n%6í®ž¡êEéoð¦ê(oš8oôæè¼…l¾edT¿=pÀ*­ßŒ¨…]ÃìŸE“?ba.œ^HFp˜Ðn8bg;âi“Vïïßñz%­«Û¦ðJ-ÉËW ÷‰Šþ¯èqKŠ†ƒð cˆxRð R+µRÄ)ÃäŒË†ü}ȱ5ˆPEê~^gð˜q¯Ññ§j9‡pﲸTÃŽªÃ0Ùû‹™NšV0µ CHLkœÁ*Y1C ñÃÂìÍè d€öadØ#ížÄˆwMljÚzó®Imˆ( 'u“†pïñ4‹p€ hõÅ(æNÒg)QoØ…±æ%íËŒñÙ áíµ½ÏËÂÄ=²ÏˆqË@pªa@¶^ÂèŒíh3b'b÷÷úô¨ uX¨(õi0¨Põo«…2ÁsûQÂFòŽ@ݱþÑ]ʘ¢z;ä2V2“Ðdt‹¸lˆ0SPde°]FV=^gaµMŸlïä6/!÷Š·xoó‡O˺ŠÐxQ…÷x׈`xöŠöBj=ˆ}Hà-‚,Ž¥üv.o%.6Έ†Hèpƒè/^8b§Ž6kÏtN÷ ˆgŠ·ø£GõŠ¨¸w¸8ú›—KøxóŽzWB«ˆgeqº/myG“BáÎÑÓp+­ÒÀÜPʦ0 š¢sv·5`*ï؇úž ú•1z¤¿ø Ç:­³ºÔšõ›íNb` ±œ3@‘†èÎ̤§fØÕ Tî'€‚Ÿž—!´n˜Î þœø@í ì,¶Ù)†4ȇî\ܺ§‰‡—ø—8üŠçÞá{¤¯p˜¸pÙŠi ‡çNG ”؉3Ð.iœØN È8öHþ3nFk™@®,aË@È ;è¾ùpͳF1í:Bþ¡¨?l4êOañ²éb ,îDæ‚Ké@ûå²½wQGìêü®rǀ㓮Ö<ôZ4èÏ=jì"2‚´ãÏ:°ÔTëÓD+ûãð7@а‰²›Ä2Q?SXÒ?Ð"±;Žˆu-Èlø²m·ÁL‘Ëæ ¬rÊ–‚ !ÍC§+? À oÏìÞ €Q­ÖK4ä-½S²@¶>¤!»ú#I3ûlSÂþ´Ò8Ù €Â ©ŸÞ ´°?Ê~óÃD­@a<¨D+ç:n»FäÜV·Ýä&Èò¤Æ{x̓KTþe∷ï—BG~=Fâd±/Wlœ8$4þ´- ?§¤&Ê´W·ÒðþÐÓ ¬ÈC då ‘7Mxê*E1‹ø—×d+xHÃñŠ·¿ãUJyÒ¹G|†1Ÿø4#ë:Ù%¯E{΃Π…£éƒ |È1!žGè#¥(BP!ÑC w²“<„°!|©ûa\ ×­ }ÑÅ0Ʊ„`,D›Æ0¤!@´6p0à=˜ÞÜm,È0,Å7¾]êo€‘È+ÁøLð35“Дü±€ãÔà]uÄÁñˆÇ:nÐCxœ×â éfL•K1\ˆ]éþCXy“ äT¬DYž Ʀ5ÈÉÂØ" ],áLÉŽ Ša by ÈÚð·{lV)ƒÓb–.\áº)Ây®Žw<&õ¨Ç>vÈz|æ@‚‘ôÂ{þ`LøÚ¶Žy ÃéÄÖà‡§Cˆ'ý1±„4ÁYþØÆPdضWÂFn‘÷R²†\Êd—0ébËw.çEä‚yD¦‰TÐô –´¥åŠb‘ ß”žË¨!«+´¶IäÃÃàÄ*’'D*=‚¡)¨8E.Ê€ùØB>1²F50…§Œ ¨z˜0Ïþ^…HjpÌ¥*S¡8Ð>ä̇rI yǬBŒ[Q1+ÚG$¢an’b=4<È*r° KÞIYØÀ:‡±Î(«Qˆ ‘0ŠxôÃûø]ï¹½–ì´§´*J„Ú@&!vTGMȺªÔ„&Ó˜Ì̽¨º™¹änºÀé@ø †!ãOsÆ#ö! ô ¬£S«8·ÖÖ‚\Š /  6eŽxà ™ðÄN¹Ç"$©J=1ŠSŒâËBhªáõ¤¤©èDØ„š#Ö±Z¢‚Hq©@è„|ã l 0Dú1„L ÄÎúD°ÐçV öˆöŽ¦À—Õ¬ºá%ñA â1«ª\rñ‹3¤$é&lèÃwqQIÌÃ0E82 $:¥‘ýñV$'ä<zày¼÷Ó’D4~¾0Ë¡”?íåeÍÓTdþ´RÆ¥¦ù% ì,AøQÈí¬Á;í„ÀØ il#PEþ?Z±„Z¤:‚²â ØÉNÄ.¶'æçücŸø¡|Ð$†CÐD$"ãAüçÊþ°ô¥Ë û%0äÞ4U;­.P:{Ò$Pj˜ôÒ"¾ß,ÌØ|°&šžÖ ž¢j“\ß@„ÜzºáÃFö±‹=¿XÁe£û nÃP|‚Ó°S!FégAËÁôbŸSI¡û¡ê&ØMf?Íóžn"Z¹2¹&sþX“3&S\Ó¢YY…`A×  ’No8=îG\!ÅNC@B¨À‚iLãŸ(OFŽå’[„oP¾=œ¼=ˆå7'þÕEìQ"нîv¿»Ýin›¿]8! ¤"Œî èá“";fu#pƒM¸ë@`‘ƒM„¿P©@òì`ûJH=–À©Èèºh !DåJÓäÁ ¥¬ÚrT'äÃ}'˜ ’¼óïzŸßoÿ—7@8´fI¡>ú I¸°2§àGÑÝénä ?È rye/}óéí@oôpKÓ „@vmÝõ¯_{¸­3ûyç~ è=þéþ{édVø™nJŒÁ 9¬ÁÜìC 0ÔÌڛĈ,Lä¹è@‘ ²y [xB6š÷-Ûõ „ðZ~þ¤?¬%xæƒ@ð‚€´_JÀ^ Þ tùEÜŒEì^þõÞþ=GðÙ ZdCÌÁ@ÀŒš¤{5Ë`Ý°D_äå Y h t h‚&h€&€B±u ²x¤Ô@´ÂúŒ• x‚¬Å‚'¤^ 6ŠöÔÛEÈ EÜ1ćO}®¨…9DP2¹Ù`ý „î ïX39²…ö€™¸ÙÜ胮ŽiÌ<`±ðÈûX_D`›xâÒ±yá°qÂxBy-èOÚ A)ÌCÛôƒ'è!p’]KÐáDäƒFЦE2…/,D¢ÿ ¢?Ì!úþx-â_È=(‹å^øHB$4œ‚=( ¯ Ä&ä@7L„& À Î hÀ¥N²Û*¢€‚L2Ü"€à„@Œƒ.ÚÛÞhÊh †=×sÄü•š2&€<$DF¤DFd ©AcDœšEÐ’!­@/¨X,‚œµÃ#DCZÆÕD ,èÀZA„<€B  @ h °@ ˜£?hÂ:ê×:z'È=lÿ¤7—?ƒ ´à ~[ ãëÍßBb„‘Då‹á`E8$DÁDn¥Tdp\$F&„N„öØÒ=Ìœ¥ÃŒO¸ƒÚ˜Â<È='rÞ@dþC ¼À \J쥥lÀ¤çmØ'0³@Ò6ì•A!ø‰/Äɹ_bAåAÆÐU†WpW~æCz¥©E€@D€FnHXªE.hŸø1¬ äŠi±C¶É ]bþ œ?¬ |^Ãå¤Lv ö Ä$D°}C[, Áú@ÒeDÂ:žXtÅdöÓÝ8D¥l&E(äjBDC‚ægÀsPC8€pÖ~çD^6¤ÎÈÁ ¤€!ä‚ d%„‚Ú”•fÔ?Hå9]6¨À h÷5 ¼@ꨀ:®£æ [PA4†ÁŽ]1p礄XDÐ̯¾+/ „SJ)ÄDýÙC8 €ÅV,«<À;¼ÃÄ\Lþ$ÝDÊ~§¬È@¸œUÑŠ`#$Y(i7|€» D<¨±=EÈê AåÜB(me¨B+Ý! (Jl¼¼"þS1ÒÖ¥¨ÚéAàÒ½:lü­?Ì)(ÁPÀ€à®îæ.ïC DŸúCðï8¶ÄÝ"Dò ¯L€óÊÌ;€ôJï@Do¡ „õ Äô^¯?d¯?l¯öV/õbïøvoùRõê‚Œƒd*CÐ 2 >@C$<ƒPAYMT=𰙯 µ@ ´<Ç €ï÷Š¯?´(0ùvCaâKþàCÿ%\Ã#àÅNøwþ` /ì[¤Œ„°¥À-%0ùr¯÷êêÙ°®š¯¨¼°÷º.k¯YX¯A °û?Ä÷îp ‡o£p ñ+°1+1[B8¯³ZÄDÀÝîî^1 ìŽïðv‰Ü:/òÖ­ò’±?ü‘ìv/AàÃÒð´; J$Ì‚Æ86‚†j²`Ä =ø@°yÔƒTÁ-Ð$àEÈ‘ˆ˜¬Œƒû «ÂòTÖ" ß›¡ŒdҞ쾜D$ œ2*§²*§2«x1D|1F,/A,oÞ.b.`AÃA€Ú±Á¸2L‚ýb]YAþ0Ü´Î}À9Dø¬Dh¡ ýëL ØÀ Á{•6ÂÓ "†w Í`EŠÕ‚­&[—DÐð±"dG G¬²<¯2«@@Û¾í@°­?\Âñ¢ÄÈÊ-A˜,Ê*¢Ž@ôÁMèÄ® *¼—( ‚®9‹<üÀä, ìEgE òA+ Y;• )ô•ø2há‘sNaòZ¤sB0l>°ó¸¸ó;§Ä òƒ̳NŸ2ÆjlÅÔ­%€ØsKT+&\+d+A§9dA·¨ÌMhI>ƒù(Ô%užŒcÃ\ô‚Æ(íÃ(Z€9+”OŒØCXÁ#ðþ#(ÆrÃÞä[›–Ý2…ˆSÜ!±~KuئøA¦ëÖôEÜ´)ïôøD*PÆ­Ep†¯%n<´€ Ì'Dh4FÈà ¼fÑÀõùÀ]Ù™£5Cdƒæ8ØÕMÅ°tØö†ê·ˆAq«?‚-22HPÍtK„ìŠFqEä4c˳¨ ?ÌhÞ(Þ.54¾ƒ˜ð°›…ë ƒ8‚3Ø™”Â,¸Gç-Ü@¤6 „*4ÇJ¨€ DŒ£ùÿÄ78CÁ C ,ŠrqË|ËfªA"íb;#ÈB^¦ b؆þÛ d8‡k8*ã À_‡y¢§z²§{‚7jÏŒq‡w€‡>HäÖÁ,HP®äjB‚:n e64APœ@€ ÁyÁ6HV¸Xð)¸/5ìQ8t¯]-ä9?TCv¸—¹86JTCœ§ æAD€¼é¨4 Ŀ݃/ƒò=Ã( …(¼ÂÓE‡ ó×@ø¹[ ¸¶¯Ù€ÃÕYFÛüˆD.Þ”×àK¸tË oÐ4ìÆ.A ÷Í‘rD$˜‡:†‹9[ÄÛŠr¤@¨C˜I µƒ¥‚Æظ1×?ÄB§'œ dÄ}7Œ‘Y!ADÝþõ9-ÌC´ Â<Á’?øÂRHB‡„§¥ÕAÈ´¤gúETƒ’l9Ñx:D˜ox¨“úZ€%4J÷ˆ8Ì1èÃ# ³ ¼·RÐÃÁ@ȃ ´@XWD¯ã7]qô@í µ‚=¤)0f?Дu Dˆ@„ƒˆKƒ$ƒ\:¶ßµE¼Ã¶§·'¸‹:˜“»ZÔrxô= A¤Ð>H$¤è ‚f0Òyµ?Äë~KÄ|ªÀ}í¤óÃÌÉþ–z0¦qÂVÃ¥;»„í Êpz;¿_Y¶a_Š•¯)¶'ãEô¨¹¸ü3B#>PÎ/tÇk{=à a5-0þRè]ž€¾cD't‚6ø;•Þe¾~ŸYÓƒèA‹zHÂt ‚ÀL­?èBM$„0ˆâ #ÓoŠ™p»àh¨ãàå[¾åW~æc~˜=~Y¸N  :°Ã#ÀºÿÌÂãÕÃ&^_j·€kCDȧEe Ÿ9hÁáž2èà ìJ3P‚P)๳ÄÂÈ<2DÀí—@'@„5D´@±-@ ð40§lþ ¸š…¤ˆ,gÎü;æ/rþü‰ËrþcG©}ìøFdI“'Eòz³r¥˜8b`ŠQ†åJ•(q¢¤FòÞ0Ÿ>Õüü™“hQ£G‰*8ZOS;MñØyêÓ§f5!‚V¯_ÁúË¥ÅE5|ü-„n¤TF ™ÅDŠ¿o9€tôt‚ÎN%DdŠ‡²žV$NÜŠPt¸cˆ§l­¢\‹;+Z$ÈQ´ˆQ#GœÃLÆ))Éb:š+ëšµHžB‡Å]›wïŽ&ˆ:ÕéTâVñ`õý•ërç`ÍmÉ…Q™š‹‹µk6ÉC#¯œxó¤Ç7ŒÙN¨w2^¦$àÅ@ïä òN`´ëþFZP‰«v<ƒ&Ñ.âG-Lɶ‚z¥1‚ #Ÿü¡°7 3”ÐCÖ€3ʨ†#Ñ8ªšRîâš[ÑE“tÉBŒàh#hþXFšGR¥ˆ"Na¢bÉ!–óZ@á “þ ŒŸ’<1i„LêóG…(ýYA0ªe ~DABjѧœ?ôñÇ:Ò6*ŠC5Q{q$‘,ôÃ8ãðM9ùô(Ä¢FÄ£Ž¦%ªA U±Ï’èjQï9ƒ ŽÈAƒ dŸfñ±R¦`Âz€ø>@9A““:mP²¯$lH¨Ï“:Z†nü¡g‰zHAb”N(aGŸ?òþñ'ŽKý¹'‹Õrâ'“0ìó|„FŒŸzéå «aö£=•ó€àH,´D =4¹rK¢ÀQxŸÃ‹`0jƒ|Ðɱ­T€ë”%hy–`ñ‡ŸlPPá¤L0(a0œdéÕ“T8a=Œ@82&¼A  ¡„~þ@GYfß‘ñ¨7«-I¶<±j[Ÿºýö¹wL"—ÞÏ5*uÛMQwý¹ÅŸ};ã ŒÔYC˜ˆ4•ä™`‘(e (FýˆS5é“ܬ¨Š?¥@éhø±§‰o¨ˆkJH8ÁLZ˜ƒŽƒœñejñ)æÞ ÇHŽ—Óg¦Ÿ PwþNWÐË“¦wiÈÃrVŒ„ACv9†·~r‰ZüYaŒöZõ#~Ú Ê£ÐîžZýÉfI½òè›0ÊäcâÂd2è`c7а0ðœÌ©é 1XŠi%5Ô°ÚÆ«ûí»¯‰óó}û“(á.oŸP÷¡ àIÈ7G?+]´˜18àˆè|$1‰0AkP¨Ç7z—Þµ@=%)›HrW’ú#·ˆGð³1zè€>´ B€ŒpŠfH¸0†?xÁŒÜc ‚k áN¾çÐF†ö³¡WÔ—“¡¹KPíb¡0?¦A`^7,Ê=´ð4TJíD2Øá€Åe`´þOy0â%‰¤…¹;rŽ‹dC1‹)cbø‚Ž|KiA0o,Áp!…4$á6ÂŘV³\¦•8Ñ#5´Sâ$DCDö&$F|‚C4IN’’uȼêçÈ“hDp|PÃ=ˆvHB¤ø‘)¦Ð„zi°ë€ì>1LL‚Œ[d;ñÅ bÃÈ*òâ+pÄEmIË<щ5¬0Zy!1òCÒ7ŠÔä5‘’Cœ ’Ý”ä%Ë•Ilzä ò‡9Ðà tb§KÅ~4°ZÜåkþðDÃhÙ‘²å³–º¥HD@Ë]â§#ÃÃÈà‚i@¢gjƒþ292½¯èépÖãž÷žó½EŽ“£~Š¤7½ ÎGE@ý§ã^¬™ê)=e (D¡@Øyè‚lÌ#$°ÝÙlÙÏŽhCmzÁ ÆÖ«cL…*F6 H(Ã{¸GÌÐ Œ´0,Ýè!EM“bS›(á&H»)ÒE= ¤cõGŒð•6¼Á‰ÇhG$T—# ÌÝ0/OÐFTÃV#ØTL†¹+A Êؘœ ‰É†²1ŒPLaJ@‚1( bìêÃEüÁ\°¦ÐÌÐ9tV±:¶"¬[%¤”¢ðã¬h¥¤ZûÔ¨±"Q‰»`Ã9þñ‡v8£Sñ|ÂlJ~È£a"G f)’]ÚóŒ•MŒ \Ë\y$E5*6PÐ ´ÒMàÆfþ |¤ÌmPFp¡Äyå®õÙlÇÊ3Ü>'FámàD$˜Á ^ðº\>‰Ó‘1šQuv¡@,bxeȨЄ¾þÕš8AR;Ÿ"V»mDÛ^À’4”%o X »Tm ˜½ZB8Ž€S4cV|_.xäq/DÉá 5`7I9›Áa°‚¯ìà#XÂr¢° ïÁ%öB åXFÊH‰µR<—ä‘î=}©;ˆ Ç-¯GûUŒÀüÀKªAùþ!âRd-’pƒ  B´dD ã2ÊÌ>2 ë%âò'“©l”&g#e=Iœå‡¡îò‹¾Œ¾`d_øPÃ.ú±B¬ÃPKˆé $L*Ø)yç\g(¹x±AE ŽE’$ŒÐ#ôq‚+LiwÊjAªz~"°1ìýÑÈaß™áÔìï<’ É`d¼rEDV„ië{ð‡/Ú€‘èl›ÛE9œÂ1=8°Xº\(‰I¡©»Ô¢~0—]þÎ"Ú\%‡ŠáEbG©ø‘B¼Ê×1ì×C}JÈY(¦²x6¶IBÀOä AØ„?jYÄeeBÆ"–ˆí–•¥'BÑÍObk8=—\Ó]'¹3mî’ ûÊn7Í|j¥š^aVâµù•#R!¥8ó´àwþà³ÄŽÂ°©›±Œ¸º4Ù2MÈý™ã6à‚ŠG°‚è‡:Ð,äo…¸y»Oâé¹ï·î`Á»œÎa{ÜDæ> ß±àä«›ù¦>á˵ |Ý£ h å:aŠ£S! öø“ô `‡VYÙîWHòŽÐCÁFðî"Ýþ&„Ã@ŠD4ú¥|à#ZXíËMn÷%ª§&(­ÆTÎ~‚Ï톯ø~ÃÀ@Ý’o+Ðù8G¸Ð'R&Å~Á ”Gø킬¯`ŽäýÚÆ# KØ>â¶N?´‚rlº4#rÀ'Þ€ô;Ú"LÄ„ûÀêIP€§ü¨² 'l³$+@¡ýÂâü²"Ä«Ä|©ôüÁ @d`ÚüAP+¿Æm§1Ü&)$Í­ †$<Ñ­ôN$"0ÔL±ù¾Pìs RÄ #z @iNg~$ ž º¡‚€²!I~‡+,ä§ÐUF@°î #ä¡Ùj¡ d.d¡La" #taɪ‘(¼Í1Ñî4ɱð$Âñ#ºÐ 'M2%Å0™fÕ>ÇÌ pœáê/„šXAÌ#UÄOŸÈÏC ²$ä!þ@¡2¨Wv¥–@ŽÀPÁØáÐäØDýd†%Ñ;òɶòQˆïå¾q¬FÒ#Ï$+Ð •ïÓr ?$ÞèER8‚Ø ÜpF!.Žà¹¾a¾.ªËéî(\Ž7l:¢Z0Žü¡ jA`¤r>.ä2¢M–ãäìdÒ =‚3¹2 áÅ5 ÓÂ2$1B·-U3%Y3Â:‚à óÊ,Ö1£²`µ”ÁR aFžŽ rr°(lPLæ,N…i"R4ì0J€àë±þHa¤¡ÐÄ*3 °’5Ä'xáPš²ÂgÆÒ÷JB%´±þ&vÊj#ºPùNR-ÓÒ>AEàÞ>‚ÀVä6—ÃÂü¡öEÖv$/‹`/iax¾Î^É#Ü£r^Ø:=!l‚?ü¡¤`ÐŒ`¢¡¡*E䶠èåAÖó Aó(â“(šlFMj>ý¡$UR-Yó$ƒ#ÔŠ#ÞÁ´@ÛŠOÂl üAØ@Ø©Tñ€ ˆ<ì½T€–Ze0_Äwp‚,ès=ºAüÁ˜¯3ü!Ð!j’%² tt¢8Ò+þËÉX#GMS$¢€5-”“*†ÔH‰Ô#" $ŽTBØþªOìeì Ìáú!/ÀkÁ¯ŽDRàwè!9y£ s1O‚FÀ¨J©è)g¡Œ ¡‘áЀí\fOsÂU/EFu96Í3KÎU5qPE·)H©5HU¤Hô`@`>D]dA¦ƒÒ@œ¨fñG œ€~à¾A-ô°´b;‚qb»J"xÊXÀ—³ f¡þ¢²œQM.B¢ˆ• Y‘Â=³Çzn´+3Rä4ÓzjZ?BZͪZ=Ö UüóJâRŸÃ@Á¢œ8¸îABU@žÀ6azíRLb(B_?Â=þ:`0„À:_¡þLA*Ï®¾.E¢h”OHÂÒftaŸö­–b5¶#xÔG?ÖZ_ó ‘À#LÖ9PV+Xhµö/ÃašáJñ´ô`Z‰àZ‚~ ,x¶( “IúÌ*túÃJ´DA¤!ä&ü âà=Y¢4gH$$7ycÓÔ“(˜5$·¶k«Õ-=äl‘¢œÆÁŒkŽ!¢AêÏ™€€€'ãÁ×N¥¼öPokCúb_Ë#T mî¢M›`вÓú¥Œ #¸ ‰j{OYâ#c3öj³¶(:×sƒôÝ8.C¶`µ¤tpäžá~$8¡àþ`Ž$ #^7Ðo4ñ¬Œ(ëŒÂL=B|©âÀa N!.¦¡úaôA(SŸuE\Îy«–=qt+é÷zK5‰‚k½ð<t%Dt‰b,8BÒ€`“àܬ[`bø âjco‰‚ì•Ï<½ q˜RSAŽ*%Yp#øC²qz+­6¨öM ¶%$¶ˆ)8':Wƒ©8%¡Š5ƒOV;G ¦#úà ú…ßè @¬Ø—ìpITLŽ,¢ŽêÎÈ^‰¢oÓ¦w“Rp3W‘€W¹3*"YÄzç9&vÊ"%67ŠPD²8‹µ¸k¹þ—~¼ø+r . †TábT™`…€öÂæ'~ûa9dX$äa \yòØÈ•3¡1J†‡ àôÊd–AÐà[À7<x‘?djÙ‘kcŠ­8ƒ'ùŠS’‹\;gy›¥ ú`öí|ãé"”`av[#ôP(MÂY1È•bòÔYúŒ†m•päVo’APÃYÂÓNäÄU‹¹O)™q•Ù#,8›Y‹%ÙŠ5Xš—£ YC#ð%t”ˆ®S†]ç°âÁÄ°dUéìCÐ_Qý!B oEat 2XŠ`œaiÙ€YšiŸõö‚ÁâFþ7¤ º#8ö$º°ŠùcÚ7<ø#XQ§»§_á言Va×(TvÞU9waIz÷ýãba‡oŒ÷@vÁ5¸àsŒÈšòjƒDŽï™©øŠñ:¯ñ:©{c©W(‰ìÄ ÈaAh1.’@ tmÓNåBåd•óKÌ4•ï#7!/.¦!ð¡Èˆ!õë†ðÀÒ“O*w®±÷ø ù®];¯š¯yïqÕ¤”zs"ë jÁ¦ê¡”1Âvû$ ½â‚l¸#®Ä *—áÂᇃ؆ز±É´UµÉŠµa»»3¸µe»6¼÷(æíièRþR÷mQ¸ bAb¡×¨°0 w™f‚K1EÀQÅa˜\a ž»LètºsˆÏg‚±ûÉ.6‰ë[»Å‘»[¯:¯8¼YCtËÉäjê*TBé”ßWÅnwÀ2âÐC6¨¼Á LI `ͤ°›ÜþÙÁ{ck_ûÇ)ÜÂÃe­! >‡,"¹Þö)‹ ¢`v =(ƒ«MªUjÙzþ¦µóì¢æ"ÔÁ2?2 il%®1™!ÇíÇÇõúŠõàÍóZÈmó’Kâ ¶@ÃÝ  Öi<üAŸ€r`øA¶H¾µ*C;Šýhiôüþ€­¡!4Ôà[Z_Æ*ÇŸ7ÁÙüE|<ÎáüÍE]Ô·ø|¨ù$bäs~ ûj2.²´Š¤F«Cë· MÂçp%â” z¸7õ¡ ’Ìimˆ×9=™=\>ý4!\¯ãÜÔKýÔÕÑÎ;B.ïÞ@C.ìÑHò(#H¬œs€º$®#Ðã>hÀ¿ü#AV´"™Ý†ÞýÙ9ÚwfÚý¡À ùÚG]Îçœ%O‰Vëç„¡Á°ã¢ˆ€ˆ¦³xòæ]!Oà vj ¼R†­­¥Ìð'no*W£þ|Ü\á!Ûþsþ*™i¾ŒgØÀ@!fGõ fRÅRùàBÙ0ÚX˜€Œ€aa~! 0"RjÀ ‰„O(RûêjþgxT~ç¯MΙËó°+åa¢}o‚ ²A¾Taæl9‹o–;Bp¥ôþ7€‹@Öaá–‰º ·ÞAÌ£Ö~QB]DîG¿ôwžî½b¼1bÕÄ!èR-–¡¹]×Xáý°Dèu¬ì$‡ïÂœàÎz¦!Ì@ˆ3ß‘v¿î~Oàcùù ·iôs^ô­ÿú±õµ¢~$å|Ò•þ!¨ vìÁ7†UC¥'†r0ºA ^!.TA–A óƒ$m4‘ŸÀ <ˆ0¡Â… :|ÈDz*Z¬x£F=@üø1B„÷¶äò×+º?ȉB‚$É 4ä}êJ`& @ú$èé'HÚvÀæ¯G,WSŽI)ÝnxùËÅE¨Ö­\ßøó qØ0®dËš= RâCŠ3rÔè6£[hŠ˜kK0ehx $éN#E¨°Ê±Šž òüÑQ¢nÃ’ fŠ,PEÎU?¼-1‚D”#vöñâãO\–ʬËzûPlëÙ´kTëЀÛþ‹o-ʵH׶ÂÄiÉ¥nMhƒ’)"sJõ°ê/Sˆ¢Â R&[-!½=¹ðëG=&¥yÔîº]müÝËr¯;×ðfa;”­ßAüHàB L”o%hQ‚ ê\EÐ…ùø‚†:ŒòÌ(îA…+:l’M (ÐãO'" ¡?ßqõNBD $Ï òt“=KœRŠ¬#:ä !â´èSŒB™#V“j4)Ö9 #%’þ(i¥•L¤`ƒ ^”`„öC¡?|ð±Ò1ŠÀ„RlÒC6* €häãÏþ¸dÙ“P> e•^Šém\ꑇƒ_‚y‘˜R¸†0Ç 2‰‡NQa «|rBNñ” ²L¸Ä@.!ÈO (A‰,çð$‹éœ@r–oäÀìA)ŽA”à ŒÒÅÈ2E:*r‘š*_ÊðÈNAÒ„“̃hÃ5žRˆLв…€€ŒdH<00¸ M8Bh Ašv A©Y "K‰Ë\äv É #YIÕ!Œ\yßýÁ÷ „Á{˜<]ʈ;œØb!…þSÀdÒ„QK½EKšä\$/¢I 3˜“¬dp®q€<Í–8@üÐR `™Ä![92DPg«B°¨ç B@èQ8ÇÙ¢yþó ¨A:Ù|¤S˜‡cçá‚w¼ã!ßÎÌПŒ™E "4¦Fñ€E©@=I,ä`C-±¸•D¾t¨Ð:çA|ÙÎ`Fò£q€@œ:š½ã¤dyGJ€¥ `\}Ú;äU°ä«Q«XBV„µ¬h=ë•ÜšV¬U­þà5Âó§Nè ` Å#’Qˆ{¤áðÈøæ*W³²õ­‹+b«Xºþ¶µ±p­,e/+YÆfÖ±‘Mìd7kYÐbÖ³š%-g?kÚЦv´ý¬%"ÀUª6d¨í/ËPÉ2Ô–—|ªo B 8 ¢ yÇyä.DH à¥zìàLp¥± B(JYYx]s¿ûR£Ä,oy%‰[óâ68P…ê@øáj¸Â­¯}…;‚ðÃi@~ Bðf‚EÕðÄó~ (ðèºÐøƒ>æ.è—Çln…›+[¥wÃ~¤zƒó‘Î÷Jˆ5F,K¼oÅ,†ßXÝç_ ú#ÀÍÕ•@àa£¹âmH¸Ã:̇¢Q—2j‘‡ºE {þØÃÆ­:ƒCx. ô´'.Ü© Ä aîPùå/x‚3µ``¡œñ‡tð Ë:$xßÞ‰tx©fò0i“Ò•"ÄË/}®@4¡‚nì€Opå(šAˆeœƒHº8œýOñ„¼ê¬³“/ Ib^ñ}GŸ':`8PàXBŽ I#‚h¸‡9l éXëRÒ!oÈpë ã:×¼&ƒ¦…¢â·x Ðó?í8Pz !Áo‹:a50Ê°ä$îP­=Ô ·¸&C·9œÞ]ÿú'–ö°=mlrê*O5’‚ЊéAûøC?Ðð¬3LXÖüÆ%­b€]ƒ{þ×ßNï¸[óir2O âY™ó#ù´áËjV¿/®È Dç5‡Å ©H`¢ú@[f@#pPnªpF ÐÑ«0 ã4§£ÆýakoœÛ÷5¤0òbîÏ-8ž·Qè¾ðJ0ýO,¿Tê/Mò¶yîŽ÷ÖJ4&'˜ÿu`X0ÁFÅ$æ‹ú¼ú¡äìôDÝ>Q«ã®Ût÷¶Îín÷;g)Ḝ‘uü± !8Á) D1Ô`Ÿ-à£æŒ·âÍ~÷ºã=òg ßirØ;ˆ7Ä»©Œ5ا»/½ø/yç:ò·®|e.¯Hdk¦:X L~|,*ÂþÁ0½ïiwsL~ò·f=©bKóëž8 ¾ñ„Þ|¯ýhl¿£×ªûê‡w×KöYœQÎä‘[T £ˆ„ –‡¢á$¿¯ÿß‚o÷0à]ÿà÷¶øëB~"”yò«ÐJFpÀLÇf3gØ4¨×ÞÆè!—nt¤"¼ oî¡«ÖS%h1Áú§‚8[‡$]·Hëæ€&hÞ ,EpË0ú€Ö`pFu„q7¸‚GèsYB(Bã‘'ͧG*w æ0Á€pwå´…å$w !|“·‚ׂIh%X;>Dþ'ð :}Iç ;h7e‚MÓA¿‡dPû§‡Ã÷h†±ƒ6ýàt±0p# ÏðrAæÔV‡#Tüöx)¨‚eh‰—؇að‡(Å„á“ "À+š ¥„1 ;µ l¦o‘Ø4“(k•¨‰™¨‡˜¨‚hÊGGýc+š  zT ͆]‹  õaH¤õfÚB°¨ŒÅ}C‹—x„˜h†Hˆ|ãgúã:° 7xRÁ kXèŠP‹±–‡™h‹ë˜‚¸Xð‰±jÙp@}=& ƒ€h€â€ÞUŽsŽ&‹h‰ù§ŽíˆY¤ ¨¤yÙ° ›Phìþ Ph@÷y)¼°Ao A4Œ'‹`™x’'™œøsñÈ7ü#š‘MQJ—hj ,FBCu¹‹<é…¡*©’ Y”–èŽdƒáC=á<´çyó– ŒÀE1Ñ…mç“‘•Òô…aF™‚EI–aà‚-rÓ'$]¶€S &¾À:?È‘æh(8––H”y™‰IÉi9-6Ön:` ÷xùH õA‡uùŠwùµ’‘)é—[˜ÐÒyÒž`=B…{°÷°˜Œ)‰Ž¹“)–’Y™Z¡‹´fýh®$ “аêpZÀv‹þ•Œ¹4’ ÑŒÿ¤m¸“’Ç™’“i’Ê9™«)—‰)ÛÑÖaSñFŠp›l@K:IšÝ™%Ág’É)žãÙœ I;MéòÐÀ˜ ¡ X•EÞIŸòxljšD‰ŸËéœvá’Ób'éyÚ°…9…° %ŸõÉ ö9¨ ¡á)ž}Ù’±sJ¶B'B° †™]C²  ”䢺4¢ºTœ½´œûÉœÈ ýéK8v¢ üК° ?ˆ5‰få0Xó—eÖ•¹¤¹ô•¡_ ¡Iª¤áé¢ H’* Bð£ð  Kç€ nÖ ]ª‡¤K*¦&Ù¤!ñŸþ˜Ò ,’3Ò Ós]©ÓÂ`Œâ¥ujÁ¦_¦cj’_e  \gdOP¢fú’¨T#ØP DÐ#!¸Ds½i§“Z`ʧKê§Ä&bA5C g*­y13bˆ¬€S= ÐtþÀ1œåôªÒ«Òt¢`©§z*¡{º¤ìõ[)&ò…p j%}R£<ƒt8¸ƒpjÀ¹@” ­•qsðXð`€¤Êy«a€°uíe—¨¯'¬-2tƒ#‰H‘,±(W­ñŠ`º­õj¯·Ê«þà^WâP¨!t“™ñÀ†9ðEÐ )Žt)¯ k;Byþ¯»­ ¶©qfD@g1_'f„IÛ´Ç°çy@ö¯Œ”²ŒT«F*±/ eR&Oõõ_•±”™Â/Eá2Ì€åà£ÿ Å ž¸ž!ÜÃŒÀ"¿e¡+üà‹› ŒŠƒþÈ[º`qQŒÆ|›«¹û±>À°(t[̯,ÈW`˹ÛÉáüÁÐÌ-¢À¶1tÖQP†É„0Xøk•¡ÁY4Ï"TÏ"¤ÉáÁÊ ÎÉ|ËYrδ WºyÜc@8YàÃþJ[J½H½H:¼Éþ,ΞüÏgøÌ[¶R£:`ÊìLjoà ¼ÍÜ<²ÎÞìÍ*ÍÒ/DZŠj,  9 JWú }`Ò' ÈÁ·Ò-½Ò­Ò1m3 §4ƒ*ðn§šºLaûÔ(ý ðÒD]ËYÝÒH]¡K;ë<%,=¦ºkpÚ\˜¬HlEnEÏh É<Ô,M×+íÕ˵Ñ>ñ\ô(hSv@É0K~\ÕÆ,”â\ÔCíÒWײב¡ö'9𠆶jj@ £yØAØÞlWÚ,=Ú¢ Ó¸<6V£;à JP“Ï—}ÀÙ ËþaÚ¦Ú¹]Ú§ Бí`öÅB ÓHp¥å ª!Ñ?qÏ ´ÜáÓÜáCÑú¼Û\MÚ íØ?°³±͇#P=}k  ÝÉMGämEæmEÑ-pÛÔ]ÝVÚnµ‘Â뙧zˆÒÄEKÛ@­Þþ€Õ¤Û·=Ý×½w¾½¡† pßz¼j©ÖýØWíÞîÍòíJÒ¬;@“ܼÀ°>á¡ðâ)®â*Žá`}X#?®m“0»³­pmE<B>Br­ Û¢âF¾â®Ñ¬fk(@€ºƒð mpÿHâ@ýxI®Û®Û®áþž±²P½„Z@ÌW~ÒÁ7à¸mÝ[ßòi4?n‘Q¼€vÐ… Ž¹på¬æ€œB¹â‡~äð]Â,¤LÐZA¬7­¸ì‡ ­öjòܵ“鱳鱓Ïg pè è)Nê%iDeŽŽbeŠÿ¢>À :]› ?Ý"ûfÕäá£ëo&ä`ˆèîè‚ÞcláPÖñnbœºA†ë-r³ßÅ—ðëßuíà5n¢NìˆÎÌj%³ïåÿú bL«ÞTÌ~¶†ŠjfŠ°‰YíÖgü ÛÞ\Ýþ]šÆõPá.îõ þ!öðžb-äºUþîgSêiºz~ª+êpÅ)OTÚþh OTãvâ¿âùº¯½úï° $¿Þ@ó €ópó8¿ó<Ÿó=¯óÐ0ô&U€0ð@@¼Õó@õ9ô>ÿóWõUoõZoóY¯õRïõSŸõ`¿õ_Oõb¿õd¯õf¯ójÏöiÿóhÿös/÷uöwoõ@÷x¿öv_ö~ß÷|¯ö7¿ó…_óYŒ5oóÀø„ô¯øMÅò“îó42_Æ>è›üp}ž/õ ùúJù+üRšä˜Ïù­OG ßê¤þoÁu_÷•_*fl¬ïú»/>Ô êî¦ïÁfn.6FÅÿ>þÎûËÿfíSûð3\Ì/ý4§ûÓoýVýׯýÍ•ýÛïýÕýß/þFæðDÕ>+vODU €\à¯D…  êéÿ]ŒþOþÝBœäô2tmÈ €š?„ .dØÐáCˆ%Bä€\ŒÀoâG!EŽ´`äI”)ZpQåK˜)'¼”ðRKK1uîŒHp‚?3 xäY´è¥‹9ÞðÓèS£,]B¥:)¦©Uµž$¨Ô&€¡[Å~´ÀÁB¼ŽeÛBÓ…H ´¥ ‘š€²þ&ë¶}࿬{_¬Æ0°à¶Ô.,qÝ´5ò;ü˜í;(ëµ\õw€9w¶\­ i¶*¤àjĦÀfA@áÊ´aΔœ[÷X‹ŽSÍí{¸QÁS-Û{ôr”eŠ†>öE¡«ó,þ|;ÏëµyBBãäÞ}Þ»z˜üÞMàz5ü—Ýñï´TÿÀýýNz'ÆKï¸ PpA8èƒ:>j´p:„¤ ÃQүÔ0!H€ ÆQ$‹.!Š:üÒºèE¨)FÏ:Ѥh„ÑFÞI‹DoIµ¸\2¤ð€ò‰(¡ Uh# P!!=2G'Ý”ó.° ­TÒ0AQÔóªÔÑ„ø©Æ¶M#(ÓÑÖ>Eµ!Rµ ;gpml-matlab-v3.1-2010-09-27/doc/f1.gif000644 000765 000024 00000035352 11422240325 016625 0ustar00carlstaff000000 000000 GIF89aõš÷¢ !!!"""***---///000111333555888;;;<<<>>>???@@@AAADDDIIIKKKLLLMMMNNNOOOPPPQQQRRRSSSUUUYYYZZZ[[[\\\]]]^^^___```aaabbbcccdddeeefffhhhiiijjjkkklllmmmooohhxpppssstttuuuvvvwwwxxxzzz~~~kk’QQ­UU·ÿÿÿ þ þþùýþýýý!!û''û ü""ü//ú22ù11ú44ù77ù44ú>>ø??øBB÷FF÷HH÷II÷MMöRRõUUõ\\ôïaaóccó``ôjjòkkòppñvvñzzñ}}ð………†††‡‡‡ˆˆˆ‰‰‰ŠŠŠŒŒŒ”””–––———˜˜˜™™™žžž£££¥¥¥  «¨¨¨©©©ªªª±±±´´´ººº»»»¿¿¿‚‚Ã파헗ìë™™ÿ¡¡ë««é¹¹çÀÀÀÃÃÃËËËÌÌÌÏÏÏÒÒÒÝÝÝÆÆåÔÔãÝÝÿàààîîîÿÿÿ!ù,õšþE H° Áƒ*\È°¡Ã‡#JœH±¢Å‹3jÜȱ£Ç CŠI²¤É“(Sª\ɲ¥Ë—0cÊœI³¦Í›8sêÜɳ§ÏŸ@ƒ J´¨Ñ£H“*]Ê´©Ó§P£JJµªÕ«X³jÝʵ«W¯3LK¶¬Ù³hÓª]˶­[· "¼K·®Ý»oå2`0V.Þ¿€ žAÁà¶önŸ±ãÇ#Kž™“0”$G¨9cÆׄA>SL”HôÄÐ$CiXL¹µë×;!AÒH3gϦ ¢Îý4oˆ»Cª>»¸qÉžÔ É2™Áí߃CWè{:Cég:ν{¥Ù™mþÓìl†u†‰_h^äŒ9œºË/ ˜Éœ6Ç =Ôú…þý‡P€ Í1D|ó%(Yf”€G™~3‘'à„¶äG *¨¡cn€Rß}”9·_…$–HÒ…¦ÈaƒH„7™Ï™(ãŒeâŠ8‚â†r RãˆÓeB£2ÉQ&°–ã†m̉kÊ$!TVÙP&«-©"H\ñÚògå˜d „¤’Zr׉l²©Ärmv(˜RŠYæ3b‰fš ²ØÉ—1BÇ ƒÊX(Fgò™"°ÑÓ”¿h¢¤$RJ‘žŠnèÉlÀå£vâ)ª€˜fªa#H(Qœ£0A:ê«þÖ©f‰©f« 뮿Í@­ ~‡°å§+oFš˜,‰ËFDÃ!œdlwŒæ $¯Ø~æǯÓ&ømÆ}Új¨Ù–kU(spÛ­|,zbœˆaš+ïU¡L îºÝ¹Dljû’«¹j¢À$Ü÷âËÝà†{ln–Vñ„'4ƒ '(ìŸï><ïÇKÍ`GÆ êË\¿ƒ¬rQïIK²qœØÊwþºðÊ8%òËóɆD%Ç{íoÍVXô„G¤ <ÏÇH—ÝÁ[gÎT¥G MÏWŸœ(]õ×8]µ|›þuÊ`§ý’cwÇÉ#H­kBÇ«öå/Õ›ðã­yIÜ‘¥€qkRƒŠùé,ÙË9wÊñ YX·F7Kv£n{v#¯~Ã>¦Ç%üýXéãZ—4©y*Í´îÆm  œ±¹c³¯TûíØ#z#óÆ™™}/?Yõ*]Ÿýù5νq¶rÝØ!s¤e•Oþý•0ýú‘• ôc4 Rñ Ãy…oBàz 5;‘A•— £dâ@Í1`W·¸ù'⃠kG¸Ç"8óŸñÇ‚ÔËq"”Ì·žÀdÍ`V•!_JÌ× Dþ1|Íç™ ¤AP¡×|ÈÄ(ˆ¯Ñ—ësBƒÄ.2Äû¹Ds¼ÿtÑx˜)ðÚ8†ChÖ ` ¢„‡M<ŸúÆHç=|ÑCî$çÆ“À1ŽØ ÂvèH4,'2ä …„×G“ü¨c!)Ó>ÈTñ 7Äâ ·¢À–È8œdÿfó?Ç"~á†^·I­pð<¯Ì† ÊÉ Jn™Ci’¥36²$„¤ÚêUËÉ(G £Á.BV.Q˜Ù›A!Š)C$Ѐ I¡c²h³-Buͤ&dÚUÄ…øaù%I‚‰•/žÇ¢é¥89dÉ`Ç…ÛþcLýL÷Íóé–ó|Œ­ˆ \'RçHØÙÏ•iàÅôÇ ãͨ2Cܬ[Eúµ9 3 Ž1ä%gˆ{dA`B+Š (PÀ (ÅP«tò<7õ 1AêÁ¹¯1œ*â‡Ð`+'S h€©XM«Ëé(. €çSÉSÇD.2uXfCô Œzd˜©@BV1M•£%‘ÕªÊN,²«Œù.#£L‰$U"LX(‚¼®# ƱQ¨h¯Œid¢7#š$Ø©a‹X•$"ŸCªM¯ˆ×2NƲA­õ¥ÃJE«Ð9þZ(" $»Â))eWíÈ×ʮĩ0@!6ÛYš(öŒ›Ð]¢È2Æ{“© DT{’঵ %`f;PÊHÞµ`w¿;^ñ®•¼ç5¯(Âk¦ò¶7½ï]/zå«^ööͽ÷…o~é_û‚w¾þÅïë àøÀíµkd„ëwÀý5p„¼_ó·Â‡z¬se†Ÿ@¸Â–Øn þzNø! k‹»éUÆËJ(‚ç\‰Næ’©}‰P‚¬*9ÞîRåwÎ…!O&¡úö%šÈŸcˆD›á„t¥òÐ$Wr2º|2I‚€-$Ê¢˜2þ‹-’‰PÆzŠöˆWŸ•’¯b}uC’ÂRª¬õ1g×,’9 ?fvŠ<ñú´‘J¦=Ó=ÉKK@ D0Õ‚&´KJàš¾>%œÎõik *f’d‚©eÖÀK¯›é É œnÍ&zü”d"Éþk (²g‘D/0 @ ò¬fMG„!œ,†œ¢Ê$ƒâ–®!u¤Ÿil’èat­±­™2lC–„®´AzÝÍj— mS°Œ×²]S28Î1µ?dA$Ÿ¯ñÀmu’‰dwU¯®qr©ùÙI*¤uMr«|“0ÉÇtMÁëVš;#… ókAî ÌØÙž˜a§þó¬g‹_ü"AÐ-lì`R  9ÉÐîtE:N;qŸü"O ö]“ {žàŽöÌM~ž›&Yø6ŽÀ‰rQ\U×CŸ÷Í7’€vD‘DÅs¡W\êSψ‹å³í¡°¯Ao 'l^š[íaH—å@¢Âê-›‹ðoy[ÎÞ3:‘»£L%“0‡Úk–>í¿GÊàIxkÌؘMÀ]&gïêÖ`æ¯;>îu!SöŸÜÝÙv†§^<Ð{Dƒ„ÍO‘]o)©¼«"-N ,âöò]Þõ »ì!sm¡äÜÙ_~M¢ýn¿Ø¾Z±.ÎOAa‰ß·äôIN½êyþOtà/d[‡ÌôAQzž ޹ЅMï²þ7Zßû9þdÖ'.ĉM HÃY’yž2Œ ¯± -×íçGïlÇq{Çv0\>}Î5Dî Æ}`׵ǃRu°Q nÀ0HðS!öçY’U…ž×|çA¢`h®á ¶2³ñØXÇò—wO÷«ÇzZ„€1·¨²ŒpßòS0ÖY]§zû' èHH„Vø #'l79 Ô)‡|#xËÇ|§…áf­1DÓ !w†ï3€3Ñ¥}º‡sÝWFÇE¢0x’Q%Ôú‚ƒþ@õ„:¡€I–~űƒBXnlx)þöÂrg 0‡îs:ÑmÎöÜq€WLYˆ€‰ qQ-qˆa¸H:}EÇÑwç‚“±>ón¡‰Ža XI×U‚3EÅÂx‘¨QÖ±“R”¡/Ž¶Š ãAH'”}rC@rjØzçA°sHyâøƒ³†Z‡~Šwu‡Ÿw‹1ŒŽqKa¢F,HáÈSãØ-€£¸N¥’ç›Gô›$±v‰‡ˆÜÁmׇl˜n8JfC¹æz`‡*qmIÖ…ò‘…ÅÆøp6Y7Bt˜ˆ9•þ À±FÎå'òQ’¡“Ù1À'|©88¯‘¡m !+’ уĸ/ò¡ ¹”©…4&¦5ƒ¤B Z¼D Y… Ño5–ÜÁ‰îh‹ðÈà'ÂR“‘|ŽáŠóQ på§@ÉSÐ8}-¸†H#LéOÃd®Ñh‘a”a Ò2>Iò8OýHw4 s²ó!²,™¢¤W€ªtw¦ãö«S”œ¦‡`©Bá¶]A¨,rZ ásÒèšÝÁ¢Á` %,ä¨!x+«qAq«~@W~ Á¹¡a82:«¹Š‰°›ÏUO‰9€Ù©´ò±¸×Q¤tD©8ò³…0¨–V2K³S  j0%&)P·BšÕ&ƒÑ³Ñq¦ [œð4O¯Çº ±µ”„¹© iþx©»¨P „µL– Š¹92?ÆV†áµQ‘ ’·¬ ›¶tdºBYMK¡0¹ÖSwúRJ* €¦>Æ«Kb Њ>Ã(,4Ú'ølYI™A¨“d`¨"Ñb–¯ û[Ö¥Ùºj, 0e0hÖ»b”#~°¹6µY±†²ÚÉæj¢ë¼®1ÇÂKM,’¸¢µK¯ Á vàÁÁU‘K}FÀ Á „åc¾«"˶fÃx‘1¯kM:„½(*º9¾0»À–ÄK|eàmySœ"–ð³VÇvdžÇfÇyÜÇ{¬ÇþȆaÆÇ6Œ"—$È¢À‚ƒƒŒJ?”6Èà%ˆ, ¯ZÒŽìÇ¡b˜5fU0¥ÄŠSÅV~&Å™RÅň¡¡‘!ÄÑo.£‰‘Œ ®\KÂrË r½Ø‹kŒ-ÅT`Cq,Ç™iz;¨·Ù3úhZ%½CdŒÅA®1‹Š„d+†™#™ÀaP…½k*¬ükÎÙSQ—ñÀŒñÈlKò ú E‡X½)½Ý(ªñ´C¾iR¢ÅUŽÁ0ÅÚZ¬©t)ªl¢£·'@L ôC,€9 û¬'q»Z‹­%UÐÒbcîþ‘éªÇñ² qÑä #Ã9âÒg쬖(l*ŒÕYZ:DûÜhË0kk'tµÌ£Í«(¿#1X 8#&a.|KOÖìa mð-_M?WhÐúƒ™ÜÌY}Ö¿ý«Ê´bÓÄn5õüM'~“ÀÖ­asÒD¢¢|bWÎ ar=×R–Ì8 ÍDiš¾YP1Ø"σ-B2½ó¬"™©ØqÕi²×ß„ýÙ·“‘ËÀ±9Ú4­×ð”¨õûÈ8¼$Š-›+ŒYËå ,çônˆàm;Ã)¸íÈ"dþÛvÜ¡}0IOãŴ˜” A Þ}¹#¦-$¢Ö €Å1ÔaÉ»ÛB­³6IG,âË*ß|ˆá½^m\¨&ç,LàʉLÑ7ð Ý3M9k„T ßRÛ$-ËóY; ,¿ÓPs °ÌW Wà 2ÝÀ põË0Y0ÚelÝd:çíÍ EZD™Ò~Ï¡ûÔ­1Ù º"d+O Ï©íémNàé[]Xq“áã qšóà­áº ¡º«£-Ò-ZfäQf A¹‘äZâ; I»l+Í=AK?ÜfŒ]cdH 3-žØþaÌÀVæ3Ž#S*LŸÛ‹¼`i>ÁIO‚ûÞ‰´P”„Ì:-RãSÞؘîÄ1ÜÝòž€¤Ö òUçž}³¡}Öu¸D’®0oîÑÔér#Ìàž¹nÇm)… ËâAÁÁÒQ¾=Ì?ÂÒÍÓb×7b¾P·ž"œ0è9±äR1xÄë³±hƒxH“åLÍ3.ÚÕQ€ä½NÑ®"•ž=°¨¾ŽñîÍ =­á3m ØL\ AåÞâ$#¿|.ãø²vMŽ3*ž¡;üM@Ô-ËW|³ËÌSî<ž1é]€²n]ÁéÝÂì÷j¨¢ìº þãGÉíâØ„’îJ“Ýëã \2äð‘ dÌ¡)L鄾 Àæ”ÁÏAå­a+µ-Ù ÑïÌÓô2Ï3 ßß±cå}ß-ªíCò·¦YFóce§‡z ßÔNŸ5'õ9¯0ÅGéÚ’‘Ô¸”Áºq¬GD§«#kß4u/ÕAæ§Ýî*òê41ôS‘O¿ÞdI¯ÇpÚvz\¿:e°ùMsäÐÊT‰¯0zÿÑH#>îMQ&ÏÝí†ê°6îëB ¶ß4½þ/•Ê$&sß-–ßC× :"Ó ½Ó ã nðeðô¼6¿ñCüf~ØsÐþ4H(-ÏOÔ\ØÎå‡ÏÜ6MQ Ô@‚ D˜PaÂCAD=„QâDŠ'T¤@ãFŽ ì`IqF …%M–äÔ0äJ–-]¾„SæÌ•38 Œ„äJI4Y±d²’'ž Òà$1È¡›'>Uè) ÑJM¡^Åê'O¹~•8ëX…%Àž˜ íZ¶`Stƒ¤ÌB•m!êésò 7ý葨ì`“R©F¼†Z»^íΛxpÏÇ•-_~,x $hÎþ)d™“qŸ´äb&Ò’]~"Ðõì”à eGÃF vB¡L*D˜Gî2TkPHPRœÔrˆ“žö62h⌠ÂÖí=1§•³Ì€ãÆŒ¿¦ )þ$§@ª“ßÇþl¢!7+9/IÌ43)®, âÄ6Q0°J¾§ÀsP2;/¤PÔ‹ ßB)d·~ûj¤#LHÁü0’îÄÊ2™À?Àšƒ$NÐ@âÀ…L´k©“þC¢“‚ˆ{h!H|pªØˆ$,5ó\ í$*û|‚oD$ zRÅŠ@Ô²-=J*À¸j 5øZh“) d§®@‚‘‚©þã¡6¹²$ó+)tI¸Œ(+8>t¯KG- ¥æ@“+áÄN¡61[Î)H“D ŃEÚóÔኴ%&¤h·DãSu ¿ŵ²9¨RšÞÅçÂCh1ã¨;©ç|$¨<=ܬ• T£=©¼%]…u"YC¤µV%s­PlB(Ve–¢x:®„§¸@³ÙH­\tZjMÂq­ AŠ`PŠ À­nkÍ×ÑFÃI;èµdT™i-.0ÒÌÒ² Vˆ¢D?€ÖÞ{Ù½–%?p8,þäJQ‚†Ù' è­P—æÐc³xòcŽãþ Â)e‘Í@È9d#‰¹$5-Û¤Ý4ÐÔ<ËeU‘Š¹ë—È[èY™Üh¯H;Ø2h>ˆ³¾@é+jEnZSŸK^IƒÝÖã›#È0>uŽs½&#Šz%SAáT„8É3dOzdc¸AÁ³VºëNhˆÁÛJ<£¾G×èï‘ëC8“ Æ,¤Å"´—Xh¨Ú$eãîtJX$Î~[ó©È”ó“h°™­ÐOÔZÕŠ-d??øšÞ?N$Ås ’·^”2_yRIJb*žQ×þC»¹3JpCÜ0Á 2‹Hœ8ûÚ7Ÿç)Ï%¯ÂȶZN°íJõ»Oº3@认ZÛIPç’× ëFø™ÀƒœC‰ôñ)ƒØ SžµF."ýXӌ娞d¬{!ÎLX#b„êBBš\hÄm=DÄ ÊŠ·I"HÇ7Dq«i~ŸŠþ‡ê½ð!<ɶò“øpfb ÑÃs¸Ò  `$’ÆvVÆ’‰ˆi\#GGÓì[ZB2ÑA¯…d'Ù''29,,ò L"†ä,¶aQC9ŸãyCø3ÍSUïCÈSð+œÿÎþRˆ9LA¢”Èë”ð;ZâçuzAÂø„EðÍò$(Kü"¢5Ùe~#K£rÈ»}%A˜ ‰–2=…v]W94 $w'šÜIÌ'Í+qÂ|6²¦Ó¶ºpáòTÈÓ’ÂJ‚L&y >Ú¤I Œ‰š*ŵ³\By¢vžDc%’²² Ë€&dSµtU$0«º áZšãBR L—0p ŽlË3Ÿ¢ÍÉÅe. I/‘3híÈ9$uŠRR”áshdI$õÆÒ¦Í0?,‰Öù^†Ò.‰V<I‚pâÝ*ªÈ“PÁ’:q$¦–d¦˜(HTþBª6­¢ƒdŠÓº’q$¨máåS.I‘D¬²9ÏQˆVµ¤Ó’œ©šŠ`˜V׃ÜÏ8y Éɺ)ƒž ›•™Yµ|bS‚Ôe-x„Êû Ç+P`q4Ý…’Qñü³ô,^Ï¢€€½p/Ë1ç;šØ¶ñneRX¬,‘"“ÓØíbÕõ!/B[–|â§ß»x“ÙJÝX7ë&§£%iM¨[É¢…µWq­D2 óqA!ÑíÒhàªAáÀR H-/}Ÿ U®¨§¯#ƒc~$ú]®d¿©°LÈGN’ pâçAñ×ää³@rñ5ÜHéž´ŠË•þo6-¿Ç)•;Ûë­&pÁ3¡ìU;‘˜íl媄¡’“ÓЦ{éŠ2X0’ñôIiOõâýa¢ HÉhk²øž¶XÛD@–í$$NΆ cÃ×@ÁR&—åkåÑy&CPoXHøe" ¶Õ£IbM’á-¹ z®ˆ€ ܈"¬pCq¹ˆAe ¦üŽ‘ê‰gƒÜ½8”HŽ[«K$˜@‹iXÝ$ƒD¡°Ã2Môà‡ÄÜ5"ŠœìIeä`*Si3a8CWR#žÇ4W±7Uï¤ÿÒÀ&±¿³8:l[ÅHLRæ–év,ôYSé\MþL`Óùi¹Òç™0,^³{ íäÌ`=©qÓÂY‘Õ äÄ °rÏ—Õ$цÒxÄ%I×@i÷»óˆ!<Å gúã`Æ›o}„ÑúJŽ‚°òÒMD˧âd &58ÿä¹îD¸( Ñq“0×'&ŒUqCbÜÝ¢Ð8~Œí),Ž*.yBþY˜so2örLZ.èöµ:]ˆÄ<ë±øú'+$sÍCÒ¦ÙXäÒYú¶ôîµ$}™Êe¯R¹ó:Á‹EiKü ¾M&[H V]«“ºÇ…nŠÜ ÉC„ò«ñvVY7ùÊ‹"óñÊåÂùÕXþ>)ïE êrL½°u ±&ýèM_ú‡ˆé³—}í;OªÎ‡b¾y„ZÝÖ®õ*“pÎO£Žß^÷´ç=ôm?yþ¬<3^I&42ƒs½Y#t‡â·³xäB¤ÓòÃ&À»î)>2m&®Ü4®µC ÜÅg·]ç*O®‘;[¾V;µp€˜;À+«¶ö1îÐ(p[ ý#‹Ó…s r“ °KŒú€À J?Q¨5- c7¸ðÞ2 Œ²åKˆœû Ð €6¢€—¿‡8ör c7ŒˆªÀ–3”89+Z¿ñð€g:“ô‰YÃÄ8 àþ¹·×+‰H°3ü¯ä-¾Š•@‹*2¿ º6 LkÂp*º øƒ Ž+ˆ¸ ; Ì Ì8d ó)$XzA/<ˆó ¼•¨AŒÀA4C;!¿Zy4ˆ±Š‰ûŠ-D¤ˆÏ; ä*yz"·+œlc7 P à•àQ±…P¤B4DyCä­(A@‰PÀœÛÁ·‹Ä\zˆƒ¯±‰#Ä lꔫHBŠ83ï@>íòD¯AP§MȸÑ#zÊ CˆN0ŸXŠÅ¹ÐÒ>”é>ð{ˆ]|¹­)EГ N4 C³»åʵ–GÄð)‚‚AH ¯þá=˜T´„ÈFƒð7Ø‹ó –o¤£p|‰Áë¤9¼GäÅ ª ,–ĈO" üµ›Ë@µ“ˆó‹`ˆ©³#ŠP4Pà», „˜|„…¼2rÈNFc `ꈀ¬«H€ëEU™€–<ˆžÁ ‹‰ à =#Fï 3S[Iˆ¨¸‚ð¾sŽ­|‚¾ð»’{60,¨AÛ ÈGPA $ †ÃˆßsŠ:Ø#—à±­›âÈ•¤Dœƒ½ÜK7Ä›D%~—äœ!°ƒ§°@ˆˆÊbÔ«%"’„¥ô‰II]ºË®™ƒïºÉÄHL´€¸ÇXÇ€Ò£—˜KĘšÚµþÁ)$É”…k“Ö NhÍ” !œÁ±l6Nh?pM¬ Œ(M¬¨=Œµ”Q¬JˆX:Ü ÙÌ\ Í€JƇÈK¬øÍ;úÊ d’¡ŒÅç ‰>Hé ç,@-!Ì ÌÆt J›"„ "Å8ËÁ˜ŠàÑÍÂy+õÄ Ù‚Hp¹Nk Ë•€£ÙˆÏ‡«±¼h §”ß‘ît¾ç¼MeLâM}£J–D»I£J5„Šô¡²c¬ÐÊÑ.ô4ψNuŠMÄG9ÐY"ÐpKˆû ßU~£”NNðªG)iàH`ƒ4]¡ö¢¨ås=j‘<ÑøMßò/ä«á«àŒsã³rŽ)hã2ê‰ Za­/q‚å§ÆÊ*–êÞÙ ùERˆXâ„hרèF7xÔ‘ ¯^Z9°qÙÙÖ7Ö‘&éE&‹JÝ2– QÚ“p-~ŠVŒŠr.ÖÅŠþO+â¹ìéص®=/óÓ¥fjÈÆŠ1Naژ㉠kÄÖ.Ô5(ƒ)ˆíU«©P‚³® LS‚•žD퇘ÄÆ3;˜æiꬑëñÊãž Xíï OxVAä!µÒi 2h§¡@»IèÕ‚À"uŽÆY¤½IÇ~l¶än Ä+™ã“xUt½¬8“Ž¾;Ô…ŠM H`äóÙ OpðtÆë +UýÊ×íÜLÖní¿Å ÉÖê<Áî“PÔnÝÂ/õí±8„ 8êâcçÐí2€“¢èm×°Kåçmá&}ë 'á æèoF’Ù6‰8Á«˜Z/?ŠMˆþ*ŒÉ ÜI@’MnÞÞ5UæÙê iUá»ì= ™Ö'ªÆŠ ‹ ±4ŽI^5Pƒ3°ç§ HnÆ‚UÜÌՀΕ0Fˆ!fñºž NØo§@X2¿ 3¿ l…Â[µ…sáðaœ¥^éaÝq¡àèF`î’@[…èò¿Uˆ@Ÿ ¯eSÉ2 v^`ìá+{V«1à9iéÌq\‰n°hH$h¾ðó±Ð“ýp×xâGGf¬¨(·lô(§ñµžöÍLX˜¹öõB•ñþzw­závÐuÂ@r‚@J‘põt'Wžsˆˆ×o¤Q<7ÖU‹„уnŸŒ¯+‰p×G2gâ«rëƒ x1øÄÙDØâ›Dnkwm]‚4F’M(¤Ö÷#% ,*ïCJ©9ˆ[·r˜aãXÄpp‰w®ÈöƒdZ>•g£?…xb/7‰[NŒ……OÙÚ"wÉXøä¦õf«ö\ú,ë–J`ßzG®Io&‹ñrzï`ÆZ?—=¸Çƒ·õ´õêÄ•¢–‡íS!·|f(Þðç“w×Ï´J_ë‚M0¡…km>zCâuMåö"´¿ŠZM =`€þ’Ïå„ uñˆô µq¬ö‰뚯§‰X„O# –'Ê4 ,ºz‡]7÷€`ÒÓOˆ9xn(A|ùè׫äÍtüÇÏ•#ØJ7Øô“ÕBà5“Àè«X[ ›¨½ þša‰Doñyº7ä›LË­ï\Ù'PƒNà‰§ …Ž³?ˆ¡ArïÆ|ît¡M2¦;Wo-ñˆ@AD,hð „ 2lèða &R¬hñ"ÆŒ7rìhÑÒ ˆ"GBœ@òäÉ‚<²l¹Ñ ,Ñ AR‰e$O:¹¼X§Â9C:ÒH$2S Œ›<$ØÔÓã£(9¡¬j•dÒ§Z·r­þ˜‚êÕ°MŠÉiFf€-8#P׸[ Ñ ’+œUùÎìhñWœÂ!DÌH¥€ÄéI(ùyzð,üå@±Ï/½M$3È„WÄÆÞm§'#=‰õÒYÒ4ÚᛕÌhÖéž.4!ׄL‘”ˆ¡°„¦—Ñæ/™ÆaJþ‹‚Ûh“jŠ:²„s\á:MèO–§¨Øɧ1>«æ44H]QTEBQB…ØL)M˜˜–Hͪ¡«ØQH·v4…|H²PD€©]é¨ çÇ¥ºÕ*Ð!úRn*1¥GÄ%Fn'Š ÊŒ9|hG(yjòs@Àg1ŠÈ;ÁJ—#)gÉrØ®xóŠ·óH {¬¶¡­àî (ØU®1 A¦å7¬™PŒ9œl5 YX6¹¢¨+f¾Š¡°Ž$UA.A.—†æꑬsCSˆ!„´I¶ /t/#]÷Ø!(AíFžÔ Pùq¼‰-fËOÊr‹þÍE/cQ¬ö"™ûh\K¼½‚¿–ýixE‘˜òbè¼NoYtÛ$Þ¢dµaËA<»•¿ùq¥‘ h„ŲSvYB·•(†Å²^ï(X!öUʇ8áT½˜x:ÖHSLßb(Äòs©=Zeªr:AüDX #‰p)’«bâ/ÉU±EqXfLÍVe­Ÿ°jó”Öª2˜þP–WsM²C ¼üe%GÎ6æiE°gcs>åž°ÌÄ›‹“‚qf9,˜U$–“ æ«h˜Iø; '"*Úº¤­•Ì„6á_įїAh“žcj£šMÌ°yp™Õ Np+Aþ ¢áâU‡F»öÀE³†ÒÎ*‹_„ÖÝ??úÍ×ògHê¬*ÁÖ-’ðôÓ?÷°oˆì¹rõì„âÍa|ý`ÓŸ{Ü_iÔ“`žÑ)`Í1 °`w ­èAœ`ÜÅ`ô¹[JRÇç™M’hŸþ–à÷åVçi¬™M Ä ¾à ¶Z >Åñ} ÆÀ ’œ.Dþ­™Ü%N°m'ÈYî!Åí B|[†ˆ÷ü Ö‰Ký9áªÉÒ·ô`OŒ^Öß!@ÙöI¡©Qa¶Xa/E‘ÖÒ_> Ï1‰ Ô_",_úAñ NÁ1ÉÀì¡êá€úᢠ" þa!â!*â!&" ¢#b 2¢!."Râ%Zb&n¢&*b$ö!&&b$jâ$†""ž"*rb)z"&^â*Râ+Žâ"Šb%Ö")¶"$¢,6"'2b,âb/¢0£"A#2&£2.#36£þ3>#4F£4N#5V£5^#6f£6n#7v£7~#8†£8ŽcåÊù©Z€Øù­9nŸjœß¡†jÀÍià;¶#éeB ¤ë©†1¢Àö?â#Å…®$Ö!ÏA(ëUIBš£AšaT¤Bê` С»ùAE¦‡DN¤Ñ]dÑ!ÊÞ&dF¦dN&eV¦e^&f6ãaf&gRÜf’Ägv¦h®ZhŽ¦i Yiž¦jNHº”ˆ(¬®D#‰ˆ(B‰¨ÇÀ ?žËTâÝR©ܤÇþ­&qŽDk€VÊ#®ØÚhD%®ô%Pþä¹@§ˆxaz £À§wZ%A”ÈZ¸ûeZXËÅì#wN'º„dO†‚jÐI‰áwÖ'Jç©„Bºà þ sŠÂnºgYÚØV¶Ž}¨XçªDâJ@m(u (Jìc*@w"h}çÑì'~þšc€Í€ GZ §†¦hUpèAx¨Bðæ„’¨–©G†ªhq²¨ŒN'ˆÂh{æ(I0 ®¤¡¦(Ž2(ƒ>h„¦ úèH HÖãh‘²çtâO‰ü'®'ÐezFz€a”†©‘†'ŠúhB¢§Ç¨§¨× $…¡Ûv‚…“Š©úh‰thƒÀ>ŠÈmJ§(ØAt6P ŠA¤EœH2„ ®8àç”J¨‘ú¢¨tD!\ªä¦k²NZŒÈ¢ K„jG©F*ª:Dj„V¦ª«^Ǫ¾f‰¾*­ŠD¬ÁpÖª®Ž)i î*°ç¬+±RÈëH;gpml-matlab-v3.1-2010-09-27/doc/f0.gif000644 000765 000024 00000011576 11422240325 016626 0ustar00carlstaff000000 000000 GIF89aõšõ? """***---111333555888AAADDDFFFKKKSSSUUUYYY___```dddfffhhhwww~~~ÿ88ÿHHÿXXÿllèhhÿppðttøwwÿˆˆˆ‰‰‰–––™™™£££ªªª±±±»»»  ÚŠŠÿ““õ™™ÿ©©ç££ÿ²²ó»»ÿÀÀÀÌÌÌÏÏÏÝÝÝÌÌÿÝÝÿîîîîîÿÿÿÿ!ù,õšþÀŸpH,ȤrÉl:ŸÐ¨tJ­Z¯Ø¬vËíz¿à°xL.›Ïè´zÍn»ßð¸|N¯Ûïø¼~Ïïûÿ€‚ƒ„…†‡ˆ‰Š‹ŒŽ‘’“”•–—˜™š›œžž£¤¥¦§§ ¨«¬­®¯°§±´µ¶·¸¶ ¹½¾¿À²ÁÃÄÄuŸIËT..ÎSÍÒFÈÊÕDÔÚNÐÝOÜà×tÉàBâçIßêKéÚäsæçíK.+öJõêñróàzèS"pà‘‚—l<ñ Á‡—™Xˆ-¢Å‹š&:aÇ!Æ ilÂñÇj8@¦Ä¸òÒH&%Ýœ I³¦ —Kb¶™i³§þO9óŸ°Ñn½½Aõ^Ø»ŸßO<ýû áw‘‰ø@ÿöè‰~\ð‡ßÚ‡à‚¸… FX‰ƒZ@èxah† þŠöÕGèµb'fa¡„(>R"'¦è¢"+^Ñâ‹4£3*"`;;žÓã&7V‘cDþ$C©¤GN‘ä’PÖѤOÂa;Wž“% :3¢:_~xΔQTå™o …™h¶©†šO°éæœeÀé„œ‚üŽžÚð‰‰MàIç _Ê„ „&ª…¡K ªè£UØÙŒå¨[‚Ó©6Ÿþy†XA¥ÀÙ¦}„)¦XíÀ©™ Ñì± é®Q*g=ºòj¬YfpðŽ¥Ävì³n0:Ĭ 4«ŽŸ(}„­%Œžj),X í¸m0ú-­þET €²¤0DZµïòÆû®½¶âûüBÔKï½ÿæð¾úòûƒ¿ýœ°À <°Á¬°Ä SìpÃOñÆw\1Ç{|±Å‡lòÈ £,rÉ*ŸÌòËÇü°ÌßÀîÕ²¡ƒ 4€P±ä>+í*PÁ´ÎVªÒ-mIÎn0+Ðy¸ÚÕÕ`}‰špŒRÿ@uмªé‚°c±Ôc“ )œ•V€ ¥ª€´Ûx“'¥zA¥èÞí£JÚÆŠ†4`iD#nÞ{14mGNçäGTn¹›˜¡9N;zU†{x5X –Øæ¬cÑyŸ·åþëËnû_ÿ$-ͶÎðþ‰ï“Ð.øíÄÃÔeñÈ !üÔº'Ÿüòb7¿Ìè¤g¥zî`¢>…_žhÍíñÎ;Q×£ÐÇNèøŠ–/}øM Ÿ¨ú×>…û“/ üìSAÿ øçÏDPF)ŠQö¥þù œ“AuýUoLàëžö¤ÀGxoB<à*ئj D“?¨k„½“a®×õ‘r(|aäb(üÑÐÔÃaÓXØvâ‚Ë¢ëcºnî†F “8®%nÂ~€âyh2ZN¼¢±²¨Å]q9ìD71FeBbBˆ4Fâ‹]d#Šþà¨E9JˆŽW´c„ðH )NÑ è#ebÅ.!„âãe0@ Є"<£!‡€HþIReÄä*‰L6B‘’p#7I Qš±ˆ“$‚ 9@î ”5ä¤/9ɼ–¿Q!ôØ?:–2äå‚T®Þ2•ÂDj„Å_z„¤%¸KÀ“ÌÛL­GD'à€Rߊ€ \¥‚„ÓnµÏaÕM(°@}£U1 ¢¸KýÍR¹‚f*i…wö-Íd†°÷ºÑÛ§Û°ÓƒržëRó$B¥*µ!`x mâZsEà'E…`)ЄÍ|qpf.Óþ MC¨t°ÑÁ :ú­ Œ“g»ht ,Œf”‚KRM=-eSÔ˜óNûºØå.—ÑŒdOMÌfFUŒEueWujU¡ºU©fµe`ªU»ª±ru¬h=«Z½ÚU¬¶U«iek\ße3ei@O êM•PÞ m>ý)ƒú„½baR¥jÀ`P l^±KHæ ›C Uï`ìL© Ί/že„¤Ì PÓh–\« `+_§(NnI8mÞ$›Y%•rúìh«5 è‘-^j â*·Ã¤ÎöЇV·í#lNÐ\çfºQ@.m€™JWI?*`A‘”þ[KPv;;lgÌ Dàk*o%B{ úÎW¹P`ï¾€#ùb·HÜýïq)`%¸Æ$Kð[¯%á¨,òoY›×¯Aq&’ð„S$æV·¹:•‘†7|Ç{|¸¹)yµ™ÞÎ0˜>Έ!aßûb¤ÆˆPÍ{Iâô¸Ç.Z承•a óÓ Cö[‘W¸«W ߌq„<Í+ ¾Tš1•ïC @­¿[že ,bsÓÅO¦B©–ü*v^LJ€Ï_¿,$-‡Ù=BrDë|çæõÄ8A…ìÜçê˜ ®_“!ˆÊB7 =8{½vS ×ÑÑߪþ)^Ó!QMËô€¤uã§ÓpÞI´º“/B׎f„Žb¹L„fœC™º4#¡R¦ VˆÁ jWøÁür±|l$™ØË®‘P{µ±Y0º„BÜhC¨¦hº›•Š”ë>k·'°1gJ½/TmÐѶ¹m¢ãI'9l¸Žwæi$ßÛºY63šo2ÙX?¦ÜŠØu$þ~7áƒ84h¥¥–Y¦…ðl>ë{~ÊÑx0¬ìg7ä&þ÷¥®Ý'ƒxw`p8ºm£”ß;Å,·œÌÑ°óˆ\ƶ비~ßúAGÑÉðòÙH\¯NDY¢þªþGüÏIι4 nˆ²p]×ÿA’O¥ur1RµFÙ6ûÔåC›uK?²•Û[öÙÊ°Ët^1“õQš?ÂïWs²Ë,»¸Wù k.9Üï~XjÕojÿ`žO¼g½{j›Ö«è(t-sIGÄ×ÁNÿ+ì\èA¥@or÷V .Ðï3[Ôׯàé¤=k‹†ð* ßÏMᢠÞÀ Â6ˆ›t8*à_W÷Ÿµ½xÆéq+ÁùJ]—²šÖ¯šU®kuë\á~òƒÿüâ/÷ߺþñ·ßb1ÁPD`q˜@!è7úóÿ~þeuÅ8ó8€Oå³z‚}«FþB¦× OÐQ‚fÔEkþÆzZ‚yħ3íwIÃV0,†mV…ð€ûvRÆqƒ× z[¦‚ö  ˜#1}O@ƒa'GP:ø‚ÁǃF€q®¡H+˜€ H„χmÃg¨G}(f}I°ƒF¨LQ€8Cµ|Í… µ…òtd8fˆVw!-¦yæ%ë„‚¾¶2x†i‘†vx‡?¦H˜‡E€‡.µˆ`|ˆ†z·Q‡áƒÐˆ‰h<ЇFĈ¡ˆE`‘wE’8kH”¨J—¸›¸¡h${XJfZCaqBñ‰j`‰P0z gˆ<68®þhH™µ(·ØE¹øŠ8@!”D¿Øo@ˆ»e½Øʨ#²tw°‹g@͈‰¥8ÔhM×ÈDÇ؃®ñƘb@ƒ£ˆä˜騆Ñ؈è@nà*¨;°¯ØkŒ¶:qpv©8@¬8Û¸lá}~0ÅVÑõ j ¹{Ü6XÙâdûˆ=Ó#Ñ~¸YzàDñõèmYB¹Œ<}'™2KGíx3¹-‰®“^à>Ô‹ õŽÓ”0 àëø ù<æ¸ «dqŽd%ùI¹”Ès“ˆàJ²”½’þú†•v°-[‰U9–Süèeyg©{b©m™]F1ŒÅ(VY Ò­•7ÒPM/}ÑB 8\œÊT-}ÕÍæDÑLm«5íÕ]s ]³»ce=Ôí϶ÐQ)^‘ÄmSÊ¢óŒ`=×zP×&=Ð4•ÎÃúÔ%,÷Ô$'‰E \¬ÍÕ¹TØ[Í,*PO@ÍEÀ¢MÍÏF}_sÙHðÖEàÚ†È¬È±Ý ³=gSÛœpÛèÐͺ} ¼í^býÛêÕÄͺ9ÝEA;gpml-matlab-v3.1-2010-09-27/doc/f2.gif000644 000765 000024 00000033052 11422240325 016621 0ustar00carlstaff000000 000000 GIF89aõš÷µ !!!"""&&&(((***---...///000111222333555888:::<<<>>>???@@@AAADDDKKKMMMNNNOOOPPPQQQRRRSSSUUUYYY[[[\\\]]]^^^___```aaabbbcccdddfffggnhhhiiijjjkkklllmmmnnnooopppqqqrrrssstttuuuwwwxxxzzz}}}~~~^^€WWº99Öÿÿ þýþþþýý>>ç ó!!û%%û''û ü//ú22ù11ú44ù77ù66ú==øBB÷FF÷HH÷MMöOOöRRõUUõVVõ\\ôïaaóccó``ôjjòkkòppñppòttñvvñ€€€ƒƒƒ„„„………†††‡‡‡ˆˆˆ‰‰‰ŒŒŒ•••–––———˜˜˜™™™šššœœœ’’·   £££¥¥¥¦¦¦§§§¨¨¨©©©ªªª«««®®®°°°±±±³³³´´´¶¶¶¹¹¹ººº»»»¿¿¿··Ó파ìë¡¡ë««é¹¹çÀÀÀÃÃÃÆÆÆÉÉÉÊÊÊËËËÌÌÌÏÏÏÒÒÒ×××ØØØÜÜÜÝÝÝÆÆåÈÈåÔÔãàààéééîîî÷÷÷ûûûýýýÿÿÿ!ù,õšþk H° Áƒ*\È°¡Ã‡#JœH±¢Å‹3jÜȱ£Ç CŠI²¤É“(Sª\ɲ¥Ë—0cÊœI³¦Í›8sêÜɳ§ÏŸ@ƒ J´¨Ñ£H“*]Ê´©Ó§P£JJµªÕ«X³jÝʵ«W¯'PK¶¬Ù³hÓª]˶­[·&¼K·®Ý»t(‹·¯ß¿€ËŠ-+áÔªWˆ+^̸±ãÇ#Kf< «Ç«&Ô|ÁaÒW„>>S¤DIôÄÐ/L^ͺµkÈž qYBÍ€@˜.(dwDJž}?D-‘×È“+oÜÉÈ«Ü6D#¸oâÂN²ž}!ö –þ‹ßz ’3‘mÓla蔆ì4º7¤DH>ÃøAßÏ¿ql$—a&Ýz†¼rB)ÂÅbC .˜Pƒµr\î×ÉH`!™z3±÷ (-8(âˆõI…(Š×†H´Ù*Êä¡«hãipXŠ<¾ÖHöÀtˆˆã‘6–BCL¶KL6$‰yág"YK–Gfib“`J¶…sR‰Xuh©¦|ú…é¦cKœW&;¾¢Ûšxš¶J o¾ÉJ€¶‘†lhêbc†LÙa‰ÑÀežnH}VŠØ%HL!Ù)hFebFŠV£–£Y#§XZé-JF"þlÆh•WvUë·’¨ ªVç&‘E¢[¬Ÿ&VG‘&‹•’uöÚd'²Iv‚Ä.ÊØ{Êf;•':fsšB&DšµT+Ó ³&fãڶ»”Þº‰a@Êåš©X¨^±{£¿$RrB¼`º"›']Pª¹1͸˜îF¬TwÜ$&™BÖv Ãä°bw "ñÈEµªÅ=¢‚AA¿ôqb§„HòÍ@©†ræaòXu0ëûðW¥šŠ¤’;óømˆb'PÌ.Íœ˜"Ûªf®"Ö‘nÒýa¨¡cÓuKRi3ÎhçÄ+×(²z(c¯46Ke#vAÚxÛËþÚlSøkc €ÐÜ+Õý Õy'"uôMáÒ-÷tÍî{¶V“˜ù‚=„ç8^7v ă Í銧¾R›Ÿïçvcw’á*~øæªçîQ,·¾ßß×*D{J¶ƒâ´îÈ{ĸïäA‹c«rèEbzRÁàˆÔ˜T®‹™×%Nç*’ÄŠvÊŠû´GßôŽ¢³Š ήŒ! ¹D@Næ‰QZŒ!ÈUCÊM¦ÈZdMôFG.æ‹pkáÓÁúiRƒ¬óäcæ…žÅòFt êª"JùÔò3TåbɘWRp&èbnyÊÄMbIºt $£HÉJƲ1ù+¦µ–LÇ4§ ¡|ˆG‚G»I•ÕtŒé¨MYµ}T ´NÓä2œ¼t%D¶)’n¾"dVþI§|ôÙ/„“1Otc ÑLgZ‹5 Ðß7ù‡É~Ò9Lk1é{æq¡û3Y'“9NyΓ”Ž9?1:²T:ˆ"¤/ayPÖà“*ÄNLµÂ;‡*cáRÌ#BQ°g£Ž1Iu·<›"Fe,“d&ËùÌǬt¨8óçŒÚ3¸qgŠ éH¡š-“þ“²IibN ‘žJ$>@p€Ô*˜®yiÕÔί¼³šOÌ©bÈÊÓ“¬"kÝÀ€¢YôChdÊV}³X¬4Ò¨¯eðúj’À­‰…e_@Ã^”«xëA!ÛQÅœB¡5 `‹ € puþHA‹7xA6ž„œdYOâƒÞd­ÝÚjäJÛ›ÙÖ¨Ë áU±ÚËzJ žMèTfºêZi•ìbl°X³RD°k@Õ +ÜÕ<µ¸Ê¢&iå4Éxw"á½lAJPÞÉ”‰E©¸¤ßðw¿ùpf\‹þ ä¿þ p‚¼àØÀ[Rð%aOØÂvpƒ!Œà kØæp‡3b øÄ"N±‰Uüa?øÅþ5Ùm½ 8³Æëí b”’U"€gï9^ôFÌ@UercJæ¶y“üéF¥®òÁ2M{gTí*FŠ»} pÉ«œó¥±¦A3þUQ±õ¶r1ucjI"là c®Å_1K#k‹[T$n£¾0“d®­"À¡= a¹~Ž”Wù䡹÷$ˆ2„&Øê¹¾“s¤³¥³.“©1Mv2IJX;o ÓòÈž±ëŠÆÖRÁ®QY‰@Š¼÷!«ðÖ\5¶Èéó¨ó¤^›:1@-e[ŠœB/R-4 +ýå¥N[F uA¯­¦»&ÓË„†´ªîM¥)ZÎN¼­² º9¤ùõFf]H¨¨ù3ÿ† ›gDǼ۠à æMîìH²Ü&$¿Ínm?¦Enø‘&]MtÓLÚ÷X»þ%ce#Éܺ,­$ÇÍTj'ÇÌCÁõWdî”c|‹yt¾³:e›|DÍvhÄÅðè[#³FŒÎþ‘Iô· ÇÏ}êÆÐÐ×<—Ì#ËtŠL4E<‚:D]&(W%¯³Šƒ#œÝä™xRŠ®'ˆ¬B4HUú ³¿$ö¶)Xô˜«_š&ÈVŽÏÏ|5ˆh‡ñ!#Ñ÷£Õ¦8… ÈC.3CX2ÖîzE|@)ÄÌ1”÷;K¾dT¤>Æ›Ÿó¹F~Io‹^"“àrdÍXy¢¤Ý“UuŒ!2Îù¨‚<5»=Eb¡“™Þ1€9NJíÐÁ‹5Œê®cÖ'þ#ýŸÐ¼+¢œÙé3®˜#ó{ w-rôŒ$}¬Ê—"Ο˜ñ+ÉBihëW–°‹´#‰§B gñÇ° b([`(€²/Ý×!ÎçPÁÇvn÷v"·°W€&Ìå¿HÖrW|dC{æÅt'* q\Ÿ$øç5eT¡Ó2\·n"÷y‘±xD¡z4¥¤@Ša‹q/?ñX6õ:¯wƒø¨ § D„aD|8ážgTÀã}}Ú·Ç·Ä%…á?Ö4îáS§Jó3`ˆƒÇ%g†q,q}g{8‚pTƒÁWì·}«ÁƒBñ}þ\±9.¸½—éY2ÔþgNüaˆØ@äQLø“YØ„Qƒ‚«A¤`‡*•èƒñ=6A‘ñ€¿”pü¡[¨(« {ã‘j“]H0>ö²‰dD'ü‘|s§&ªpt0Z•¸‚$á‡]‰¯Çr&H7¤˜‚IÁŠX‘+m˜‘Š+asÎFcQq(‡£H!³…Šª ‹û66¡5:èACF4÷ñ~9wŽLW{P!Ñ{0ÑŽäqä4‰M…;…ŠÀ#8ÐØDk–†.7w—ˆ¸ì2ƒ‘‘€’ÃŇ/QoFõl!‹³w¢’Ëö¥þ#498Wx5¡•y’ŽÃL×XŠÎˆ^Ýè *“)^àq%7 ;žÖ0N® ‰9˜d†±ÐÁ ó²2å æ±’×2. HøzO¹s ‰uHCéEÓx‡KY–‰á _©—‘Ã%á‘ÿô†áˆuoùm‡ÚxW¢ŠŽÁ*KДˆá ,B/±†‹ÃDÿˆ 7—ù’˜¯Q‘玉á<ô¬’•H ‰œ”„­"%xx‹2“!Fˆ‰Œá æF'À MÂ8i GŸ(‚h•´ØÑC?Ö%zØO„Àâ«þP>/±ˆÕ‹´©øhúˆjG±‘˜3Ô§ƒw ­ q2ï 2$‰ºöOiž1žñ„äAšËvŸŠñH°­±qb…éAN UÉœ,Y!ëx{òÈ3™š ŠeÐÕ)9,qœM„’ÐcõihÍI!eÈtgIÅÊ"zµ9Ô C<)K‡üIþ9*¯¹£²žˆá<Õ :x~¹WGhSSé*ˆ¤QžÐS›2ØxH¨*ŒzT²„,щ6'QŠoâIˆ£©|\´K²‘Ñ  (GÉá¥#ኃ)óéþ~Z…”­z?7ir£«a fJ¦Œq,«cT˜ò5“±œz)’•]‡†‹aViƒ§¡…‚NÒžo*šp)=ž*ty§VÈ1?,"©4¥ 1p'EŽ›Rœ¢x‚¸Ég9T€š*3œ^¤¨‰?Äš·“P´¢¤…Á¥GÔu^Dv꧷«Ú  a¢>t¥’1|AB®5Lõ„¨­ñ£—‹Œ.Êq}ÎC™‡Ã­±ªÕÔ…®˜èx°ÆVwà«d–"Ï)—jRm¦¢Ú,2£8š¬ò×8å<ƒêQøª!! »P€ áe²áþ­)ÙƒjrQùMš³«H%!Z6•Ÿ„ÊÛ„/Ðj—°ù5llõjkÅY²F¯çÄt¹B~ƒmú ¶HÞ™Læñf™ ´ D­ò ®5Fœžæ²pð;T¥‹sÔÖ˜:Y#ªá4‚Èêµ €°ñÅVX¨Vm¹h0CJ†F‹^x¨ÿÑ «ñÔÑ-])¬ØwhœYMžÉ}|~Kñµwp°Ö–µ°VÝ“g?Õ#BuˆZ"cCx¦Á¢ @z“ }@e,‚M€ã“£ Ç >­Jr²:ÏÕ«p ^$p²h‡‹¼®…¸Nû´þ—K[,ð8> „*¨ŽA$¡®D¨W›Ç»H»VÍ‹5€ë›5%0†=BC7Öbù[búËbüû¿0bìb6ÀûkÀù•Œì_n‹zËv‘@ Â_cg'qE³À<§¸bµ°£Tq“pÀ–cBÐ<¦îë¼ AL«š•hMÛ$Ši:‰›»©—¥µõ—~Q!¦ÕÄ®¹i°k0|± €5¢¶Ìd¤H"Š1/VøÛ‹}སšœŒ±¢Ô ¼Ŧ®a¡¥q„u<‰»ÆI&5xþÊgéSÅþUÀñ‘"ü?×ù±}Úƪ ¸à¹bÅ裱‹´ªú}P¼±§É€T k(z¹ â ‡ ?>«~Œl±0l$À>’<@wÌ?Àš\ܧüäûIjÜu ÀÁ›á´–«Q¨,:ó««Ë$>ÀÄß´¬zì”êxüItë§õZN„Éš|Ì!ÑZ°Gs¼EÊ´Z&Õ£8Ú A ï$¢–KhqÊà#µ­‘­ûiˆ¸+köË$Æ°¾E±Œ£ßL³  ©câžÊuB̤²äw€°ËÉôÀ­1ÒMüc]`@žÕ ÖFV 5 (:Ð A1ŽñÕèƒ&H•ºÔÓ±˜a]qar²àgAwÒQ4­!Äúú³aÏÌó ­Áز7¡MòÒFæFb*ȽJ­åuª ûÁV¹ºͯ‘͙ͩ`M>‘ÎRñGÄü 9ŠÌ™´]„[}sþN—¬˜ú¼Ïag ¿‘Öí…^÷ÙÒ9·¸¡Ó‰q³Ëd8›¯aK]¬¡Ý1lîW؆M[;|©Ðcx±'pBµÒLÑÀ&Æž$Þïý„C ÝàÑ›øMÔÆ;°‹çL׎1¦8Y+•¿˜« Û‹ýÜàñÞ½žÊQýT§Œ¡³¡¦‹àaDâ!ÄžäÓ¯± þô-M½«PÇ;ÑÖ*HàI3)m±À-ÑÐMAh­8«Bî$´¢--žª\µÃW+n¨òF˜â¨´2ËšL¬‚±¯q˜ØÓPä¼eZµYU÷æ qÕŠ äCdw¦yสÐiÔœNÖ¨þ&þQ˪ڕþBÌáÒlëÑî8×¼†¿Ùî&E-M“>ϱ3€Æv 6áãŽä쯑惘Ü}Gλ1Î=qzþøÚ ²_¼­Jè-wÀîXï&JŽLžKT'è†þÙ#qéŠ!‘$:0gÞ)÷ÜÉ¡è‹n5ìÎb~î¯åº3Ý ·‹1)%ñáŠ-Uû $Š¤펤ãÉ!ÏèNÞ 1æk;Tó^p{òµððXâØüÅgx”­Jâ.2ïëPÑñ&/¾P O눱 U4öù>È^Jõ–DE¿« …<ó-ê&¿êÂìè&Ù&ÁïÞ±1Æ=KÄê OçLâ× ¤…[ ·­iã$·)/÷‘“qwGUŽOí!AE±pcmgw zzF O2.ølÛnœcÏ÷‘¡r²žL‰L¥,ø Ájˆ8ûŠ¡øßþ÷+ÿ æ dîõ-s6Ð=`1ùO™Ny°ú €%ˆ  ±€Ž|vËÜ'ß^µžô‡nëB )qWwY’ÛäÁÆQKà@‚ $HáA@$ÀÇBŠ[z•QãFŽ=~©¡âH’%Æ2™RåJ–-kÕÁ¨Ñ ’KEºtÙÔÇ)HÚ€TèP¢E‹¢AâÅhP=8+*,y€¬ƒ± p*ðâR®@i”ÊVìX²e ‚Xµ ’NOµ0K‘˜,qÕUï^¾EY­mÛW£¸¡’°aáa_ÄÌu•¡:…-_ÆlÖÃÆNk9Öþé“¡KhFFº+R¥©+><2ñâÆY·ª^úPônÞ½–¢±±MRŽ'PúØc§ÇáSp?‡Þñ/ÛÔ†"YŽ]Q€ã­J{;ºPÝÈI‚5Ÿ^%ÌHnÄ žTpÓ7y <^?WÖ©'OÂÎ$@xá Ä1ÈöûˆRÔƒ0¬@àh¦ü^yd"õHû¨¿ŽÞc0Ä ¦»P0º3+»ŠÛ”bA„1ÄOÄ&{MBw$‰ÂLãè+}ˆ$­Ž:C‚•mdÒ#S£á2c +¯Ä2Ë–jl2£òx3L€ãˆ¸<ˆp•@š %»ì’DÜNþÒ¤,ïÄsËáüÌUÄÜ‘=<ñÉ-¸"äÐ#»ÈK88»ÃLÕ L.»¤@1QT=5êìÂ>B‹4z1£TeÉ#£³R/er•H4ì4Wõ6‰â85/–5?.‰5Ú@¶-’EVIV¡s¥§1 ë¡Î’fCˆ±ð 1’Õ>uW42…k-£UÐÔ„S>rÅ´TW}6ºá–pVµSª•U6ÅÚnÛx_9áÏpÎÌÔ¼HõÖ©¤y}ÞU›NbÁ !¶Š&Á³ãüwT1Ñ#˜7u7‹ …ÈXÂO=BU#‹'Žl&,¢3DS)¢¤ã;þH0«Ç&þ¶d£ÃâUfFÏ´t9E×rvfšùÚ°è2µ±}/£µK€=:l–J¡k£4nÞ¨hõæª'7#ž·h‘xº(³i\-ëºÉ:;p—ˆä¨ŒT Ù[ǤŸNòLâV-R¼Æ»³)!ú_PåQ`ÁãÒ ÛŒ¾¸O£¯<×ч=9rrÕ†C¢ôè8ß1Ÿ¯ùß›xûó¬„íh ˆÏ­¥÷ôÔjQO\ I©# Är;oß`kœúnÒAß·¯¨:0a¶'Â/ГÙDmæ÷bÅ´r£[åéq*¥!ë¯_ýÑHðå¾ÿ‚ZXG°Êjôù¡þ µ¾®¸ H˜Ä¢ ªàÍ%Jœž¦¦6ÿùÁÂh:’…éA™W½x”µ,o?†¨`¾pr€ äî_ãQ ;Ø’qeDk`Z-t¨#ƒJR+ Rð8]g~YÑ–mò÷(ïíЄÓÈ%Ò—*J(‹ÃÏZd‡D (‘‰ 2Df.·’ÔÆ)Øëí¬ø¹ÐµÇ' *`®t¤LŽŒìzàm¸'º$ª¡ãVÔÍQlÀÛHO0‘!媇ì2Í´þØ®åŒ!Z„§w;B¾qŠBÜŽéÈ’ ¢a™Ž w·Ê$U£Û&9cA2)”…¬Èíþp•Á¹ˆèt·#ã±r!DÝRžfB¨w@¡{¹IØ-aŒ""„(‘ ©Ž]B)i´—½)u[…ÈÂUG‡¹SŒdÅL’"AÅOœ½™D•ÌyÊG½eGŽè$¹ÐRL×Ê{f²®J\`—"ŠŠæ@X(äÎU̓‚©‘ØH,ÓV²D1'#¤(ó<Ñ)¼ÊFiìhfá!4ä˜9g—ª¡!–ô -uœ45bЕGux¥K3Ò>ONmèÉZ°ðI8e0.kDÈ*|P%H@¥ˆ•BêºÊ\óáÈ;7"ÏÅBõÔH'sÉþ*Vl¢ \‹lÚ¥ ³"€hãÏ {­Ÿ ¤e•ƒH Ï=H ”mê@0;&Ízn²žílf/ ÚÑŠ¶x¤=­i7+ÄÐr¶´®M-lWûÙ¦ž”uÛ| þD[Ù¢v¶­eíkƒÛZX–¸·³–ŬN( *Ô+:>/du-é;tŸ•Þ÷·Â,(xꃀ§$¥°Ò tF‰×v‚ì£6çV]ÑsªP bɾH,œáºhÐn_<Ñ dyAŸ×½îÆ]×åA4`-HDˆi;`FÓ[«^<á×1 ë ¨(_¾!´ºõ’IuÎRåXâõ ë„u¬aþ£Á ¼D0‚½ÐáMxjáÌ)Eð1„ )Q«Eâ‹Ÿç8°\å h DÒ¬°XB¶µcÝ^A׈ë‹Ê„H0aÝ) aü„åtÛ0†;9Á^@C&‰æRÊ° ÁÖA¢ÈØ eÌžÂÚ``À.¡ t&Êîv‹:hà| 3Ù7#ü¤nÅæƒ>¼)yÖÍquÝœã@# ĺüÈ)(-C!ù FFrckÔŒlBÆßxéM½Ü¡eÜ+NЇnÙŠ  ònN슮4pH ³PX ]úº;nÇ}ŒjŽLƉwFHDrD _˜ÏJŠ+"u—þŦ¢ƒ¢èí!B숭. NaY#U[‚¡‡Nµ+îD¾mn Ūx°IÐ ¬ díuŸ?² ^ŽAÞQ÷Zjå_óÆ©ƒ+ÁPŽìËXÚžšÌ—a.8­á»[)‚X<M%Iþd•Èë®`‚—P ½ ‚¨ %`y\NüŠ…šÎʺ*EÀnN³œƒ{!bÝŽ•Ìzõ‚™§Ô«ÇùŠ@"(¾þ\,4€Ò¢Ø êêñ÷}§t}ŽÖ[×Ï*ˆ°#'™õi{dôêöT›7p  ûRú ÇÞ\@™þMqF˜w0Nvözzhå Óþ]ñlƒJž²W¤ØÔK(/ˆr$R©’¹@Z¿#4Ýó]â_èÍs)Òß%ħ7~QNÑÖÞD!>€ý^î0{ÌH̦cC«6hÿÞFÁ· „2§£§ïã{)Ã?3îõ¡à{¡Áê-sÀ#­E‚}ß^Êæ}&B„“Y,8?¾h—ô1Š p>™„a•U» [Rš÷Ù<î¹7þ³o@KŒ,@[ - ŠT˜U¸|[ íZ¾Ýs äéˆÂ=ù úÃ@é¬q˜ðÀñ/Ü Š89³Š˜»½P¸;Ì eb6$1M;ˆ"Ì•š«þÁ¡l‰Âó­P ´‘>:>….Ú‘ 6$üˆH0Ãßé¼W8ŽÀ¿þiÀ+dÐ>Ã3„âCé¤" à·0q¿¾˜™;𪱠ƒNëˆ2@˜{… ôŸ=²Ãñ˜Ãð‹ÓèC&¡ªçY*¥ãRxÃ¥šıÀ:{ܪrÁJ¤ÞàÂÞˆ)¸ õL½éƒD, ”;Ô ©1A*d‰?ÀƒúÀ/ú:¿³9YL æÓ99‚äc•™Ä¡°¼Ìp>È3¡ 0å Ÿ9…% Zˆ… ø-z9¤-Ì¡÷“ÆÈ?L„ˆ‚1l’O$Š2þ “:Ø¿¾€§kC•Pˆ”2¸‘ƒ—«G¾ØZ À%c^ ÄTLR Ç®ÅW¨ƒ‹¬+ $xb¿òžˆD Œ¡H¤müÀWpy,‹X8ÁÔøÈW‚”l +d ƒµ0œd1KX§îcI®0›\‰Zä¿ÛtœÆ°@ÆZ€¼×)ŠA“èƒ ¤è‚ 1C÷ œ„\J¾À© œn¡™\žçQ´Dèà 60ŠX¬•ˆ…2"6ÈUpÈÏ‰Åµì »ÄGõJy ¥Ås´&z M0ŠsQÌ’ØAP NÈ‘xÞüÿRŸŒÍƒÚ¯ŠÉ˜…Þ€ÊÝ@Ω±¢XAÜP™ŠKà%©a…u‹)®…h§êƒH€Á qþNGjŽ„ѹò Ô b¼Ž@G¾ñ=UK‘ãÇ#)½3Š´HR@J˜„U`4 Svi¶ Æ’F¼RèËjdÌôl’+€ ™Œ‹!ÂK©ú´MOx #íŠS°ÔøQ¡°‹ÌÓˆƒÌÌGrÀ>õó<ÏÓT&Yw)ŠÉ³ŒB䯌8£EÁtK<¾°R ˆlQ£jQ £ÑŠ\dÉÅ¢èɸ¨ƒP}PdY.eiƒi A½Ð«^üÒ½¸§¨$ú_Õ,VæQ’ÖéÊÜd xROû Y5ÀHUÓ4 »8x¤¯Ë Un}ÁóV¼ÕxþQ廀¬¤y=¡‘6P Y効éDŽp×¢ÈÕÛtÈð—*µ×ø¶R•.e’™x¡8 XøX‚Õ6àDÔàÃWM 9™«¯$ “=­ØŒx„‹8˜\!$Á5 X¾5äSÒ'¸‹VH mB؆­O¥ÂV‰åžXH¹™5„SÈ×ã þ/}”bØ­-¸€ˆ¿ôôáˆÙb®׎xÝb&FLX®ÝJæ64f$‰[  < ô˜DðÁ]bCþ¶½A.Šêýæg‚ˆdò¬`eFŽÕ¥ äñ8ÀR–˜¼0ðŠ˜{Šl~˜ŠcNêf0e AQ;ŽåŒe批d&‘:“Hð€ˆZ …aº~•\Ø ˜(Dé”HâN™`&nËr–å­»a]Ö \ÚÙh«ãM‰ \‹ö!Nü@3‰lýçg–¬ƒGfµL,éç0„³2Š,di3„îˆGS°¬×Œ€j·^ Iímöˆ<½êYsåµÔž2žÞg9…d½€:„ Èëv´šýI¢¨ds DˆNQæ¥`vºîI¼k@@ÑÑ$g¡&éxñ€>èŽþUЀ ŠÄÞÅÊéË RxlZ Aèê„uCkœáßX'Ù%O»ÆX¯aè8Vƒð­¾…*& ‹:4Å33¬ú/¡8ºëèµh‚U6\@îŽøÁNX5ÞÚ èiW5ŠM,ƒÖí»VŽýbóÀ\ÌÐܾX¾fr1Ã瘨ŒÖˆ-‹øÝÁ„1X‚)`°Èé.DÈ쥄`ú~â ´®ÆÉ{9Ô5ê2ŒŒ‹­‘½îhmë¥$A¾ÞUx6 €h¤ûÞ‹î% áŠâëX¾XÉŠÖŒ‹a‘÷.‰žÖ•¬–Å;àꮶþŒRh£Ø€(!ïŠUp¼‘Øo¢À¥BÖ䡈Ÿ ¸ Žl¢ë†ï–FXe±¨o— ËcrµYÓ»!¿@¢€Çyë¡HY¢8bËà)‘BðseÉÍælËzcŒÂ2hal“°q¡0£= >œÉßÆ  Q$ôÍTÞíÂ8¾‘”óÞ [!s„Í$Z Á䊌ÑSx_œ8U—¼ì”8£ªîz”ÜÅ4‹íÐ!R(«xtýÀ!•0Á¥P¢åÜg­²MHó–ÈñðgBnÑ]KÂT5g ³ ;B`Éd µd!ŸøÀ•ö¢HñÝ8Âçþ°umGˆI0tiœð݈8…ÓD qrŸs£H'âê®É*H[N.[$xÁL=[[ µkO 8ŽÕ2.®„-œ„Ž;ȯŽ7ù’Gy„?y•Oùï".RDÌÉà—§yßò­S „œï€Ï ¯ £òÄ ­hN¾0âôÁ4ó}2´¥°FÖ|ß 51P>´@Ì: „oç Ÿºd‡ŽWl ÐðEîWÀ('@k³Š7Z£ÈôàôȘ¯ßq+,ZJ«iËØŽ®'ôr¼ ÷ó÷< ÞáNh2RÃäŽèÆãñ]çO©ŸnJðÄõQ‹R/ˆþSª‚Š»w 0sŽÈfWà„A[„93½ˆwó¸vÔÈ|{ˆñ´ÃÙÇn±Xô‚hôÏ—ü®8ø*;q´ òŸ,ýã¾JÚ×ú¥¼õ²w–hó€s‚{Üò• 5Kï NÓ‘R0ò¡øs§¨ö´Œ¼ˆ va'Â&ß©rd2w}7$ Wèð @÷Ô‰p£@nÚ—hªˆW,hð „ &Œä£Ĉ'R¬hñb- 7r´¸êE^¬šØÂÔ ï êè²–>*gÒ|EcÒËœ:%jU³à…’;‡-j4§‡ŸJ—2-¸ÊPŸ£G5J­ñdÓ™ ¬FŒþ¥!kSP-¸Z]•t©¡±dײe«aظr ÁÙÖ%Õ»C_ œ‹ðÔÖµ}„ø¥éA¨Þ¡u|, JqâÈ:}ø,l9+ È’)æÝÜëe‚w¢²õp*t§=ëQ™fÚµ±XÓƈèêÜ4זعwE¾º_˜Ívá5¾ÑìOÇΧË>¡ü:ÂÔ3n§xnnÞlADÂþJéîû” )V=|£ ÌÓ(ø„øA‡^Õ§Ž^±\€`úYän)" rÔÂkõ)§Öt¿uÇ_h'h&XrÊÙ`—ƒµPGJÒV)!&ÖGàI¨›!=lgávÂ…'™kþÃrŸŠµ´`B'Ò#‘­bËÊA5£Zö€‘A§ƒEÖB^B4¤g¥•g%™ÛMLêçda4œ(™$¢ÖÇj\Ö 'D(ÐmºYd §}‰š Í9G#u6†ækèiÙ)Ü)%Èé Ș؆‹U‡L{†vAw€NGæ\ïyVŠ—…y¢¥QÒu ™Ÿ©UJ‡šZ6)už:ª\u ÉZµÊ埲»h©¶ÎÅ©˜ññWqµµ©\‰{­¥ó)ë×*¸VØd_—ñ(™€s6e"¶érYG¸Û‚eH±Àé ܳYïf«|×bêÖöª¿.!ç®þ\}´Ô,|ß]VÇÁ½!Ò^V‘Œ0Åð%KðRa"¬^½çÆJ{a]ðqÅ%o§-ÆMi×)¸—é»]Ö)µ`¥&Û \¤|®zóöÖñRu³R&Ö|3WG#] "„í¼…¹¶lY]ê±çbJB¾¹t€\[‹ÌOÿ4qm>KÛî\„ÂW‡æÊ {=7k(‹=“Ð,ë'¨_d· ÂTJÓ}øZ=¸}wB†üºñ…hÇoƒ€\@-A x€/Ⳍ3ô%=Oí Ç áhIƒ’‚‘êžu¦çGŠ®RŸ¦)9X~ÃWJ³Û~¼^‡íγz; þ|VÂ#?=×y.¿PÔRÿέ!QRÿ=â˜^¯Póo§ð\©§k8ø±4­ã´Ê½ð™MÐ4½œ.ÉícÔ?Ýì&¿tÏ~§‹ àü§@¯ p€ç1Þ·¶7—¼-°‚&óô®—JP.,± +6 !`m€jóÝóü­tÕ.„i!â.&¿•¡0a,š s¨®ÊOQŒß²rŠìé°ˆ‹¢Œr o™:èKgÄ)ºé6I|EŒâs?Ö±)fò×û*F¤íŠu€`Eˆ¹HïNtáé&Ãå©ï‡6œ‹©¨G.YÏ4¬!ÇnH î±*bQOÈîtþq)>@£!#©žø9°¬Ù¢gòW"ŠáŽŠDÞEGDí¥0.Š”$*áÃCÑQN‹à*¡Rê—ÊYªƒ4˜~0¹™FÖÄ?R¤%0{C ÆŒo³#ÇÖŒ‘KÍôß3‘J±ÙÀ@º”Œ&rJþ2Ž‡Û×ø,yÉ2EœÁ<çZNkˆ—°á:ãy+^oˆìì S„P:yòSJHº^+]‰Ï¥¬0`[2â*á¿U>­ûh)™’ÀŠÁ3„”xè÷ú¸» ê'?ý‰=)ºÇI@Òvˆ\^ù¹¬°bBØãEýG ÁÝm”Mü” RGŠÑ`¤„Pà41fNÏ\3þ2¶lŠ  Ô¥2µ©N}*T£*Õ©RU©¨*V³ªÕ­rµ«M½ªWÃ*Ö±fU)¥„"²²µ©Dè'\ã*×¹Òµ®v½+^óª×½òµ¯~ý+`+ØÁ¶°†=,b«XÛÝSüHHFMä Á±‹]à,«ÃR„äÈl`¸ÀXdö²þ+  Ùfö¥©ìt8‰ ¨Ö´ß„js[DÜæ"À/ø‚ÜÖÖ¶Ó¦|PÜ÷—¸}Aáƒä.׸ԫn%Õ•("v­{¼ðZ¼ä­àyÅ‹¸ô*м« !{Õ;·ø¶Ï½à}¯|±µýò·¿þû­}Ág_8¿n".‚uK‘SO»ª›„wïkàñâׂÍÈs%Š‘ÇYx"ícübléøx I¦Øã¹ÈF>2’“¬ä%3¹ÉN~2”£,å)S™,C®2–;we£l9Ë^>q‰¿,æéuyÌffËri[‹I¤Ö»rˆA[ ÁOá-.‚AÜÀOÊ%Á™}”4Â5Ví"çáÂyΡU4~‹»_Õn ¢Qn›ºÛgAsz(„ID†ÛâZˆ/0N,R»iRGzµ(–Ep&ÚÊ­Ó¶þæ¡õ ‘X,—ÒÑn¢Y½`W‡ù"ˆ@~2|ëeãZ³´ 6¤!½Š[Ïw&öPR+„@ÌþöEm¼^àc¼¶ìµ%¢ãX„¤Ðà~÷°#B[Œ "çÅs«Õ]lŒ(Þ„7§ Mor›ßñŽöNº«ZK<àÎv·¾!]k¦ÁÙÞïUxÄw¢ÝÊn·áð÷DÈ­ky تõsuc¬\+÷»·ÀC®ÙÜžzת6Nf¡ ܦ|µÃ¬¨¦…20üåbyÆYÝæIß8"Ég½«]‰„D«Qn±M;ÕNÖèQŽyÒk í>ˆ"†NçX¨9"€:‰ž9~‘•_ºâ^¿{Ò³o¼ó#e¶%öÞ÷ÁË\/B4áßì»AŸT<äóùÉK†¤”ïH@;gpml-matlab-v3.1-2010-09-27/doc/demoClassification.m000644 000765 000024 00000006036 11422255415 021611 0ustar00carlstaff000000 000000 disp(' '); disp('clear all, close all'); clear all, close all disp(' ') disp('n1 = 80; n2 = 40; % number of data points from each class'); n1 = 80; n2 = 40; disp('S1 = eye(2); S2 = [1 0.95; 0.95 1]; % the two covariance matrices'); S1 = eye(2); S2 = [1 0.95; 0.95 1]; disp('m1 = [0.75; 0]; m2 = [-0.75; 0]; % the two means'); m1 = [0.75; 0]; m2 = [-0.75; 0]; disp(' ') disp('x1 = bsxfun(@plus, chol(S1)''*gpml_randn(0.2, 2, n1), m1);') x1 = bsxfun(@plus, chol(S1)'*gpml_randn(0.2, 2, n1), m1); disp('x2 = bsxfun(@plus, chol(S2)''*gpml_randn(0.3, 2, n2), m2);') x2 = bsxfun(@plus, chol(S2)'*gpml_randn(0.3, 2, n2), m2); disp(' '); disp('x = [x1 x2]''; y = [-ones(1,n1) ones(1,n2)]'';'); x = [x1 x2]'; y = [-ones(1,n1) ones(1,n2)]'; disp('plot(x1(1,:), x1(2,:), ''b+''); hold on;'); plot(x1(1,:), x1(2,:), 'b+', 'MarkerSize', 12); hold on; disp('plot(x2(1,:), x2(2,:), ''r+'');'); plot(x2(1,:), x2(2,:), 'r+', 'MarkerSize', 12); disp(' '); disp('[t1 t2] = meshgrid(-4:0.1:4,-4:0.1:4);'); [t1 t2] = meshgrid(-4:0.1:4,-4:0.1:4); disp('t = [t1(:) t2(:)]; n = length(t); % these are the test inputs'); t = [t1(:) t2(:)]; n = length(t); % these are the test inputs disp('tmm = bsxfun(@minus, t, m1'');'); tmm = bsxfun(@minus, t, m1'); disp('p1 = n1*exp(-sum(tmm*inv(S1).*tmm/2,2))/sqrt(det(S1));'); p1 = n1*exp(-sum(tmm*inv(S1).*tmm/2,2))/sqrt(det(S1)); disp('tmm = bsxfun(@minus, t, m2'');'); tmm = bsxfun(@minus, t, m2'); disp('p2 = n2*exp(-sum(tmm*inv(S2).*tmm/2,2))/sqrt(det(S2));'); p2 = n2*exp(-sum(tmm*inv(S2).*tmm/2,2))/sqrt(det(S2)); set(gca, 'FontSize', 24); disp('contour(t1, t2, reshape(p2./(p1+p2), size(t1)), [0.1:0.1:0.9]);'); contour(t1, t2, reshape(p2./(p1+p2), size(t1)), [0.1:0.1:0.9]); [c h] = contour(t1, t2, reshape(p2./(p1+p2), size(t1)), [0.5 0.5]); set(h, 'LineWidth', 2) colorbar grid axis([-4 4 -4 4]) %print -depsc f5.eps disp(' '); disp('Hit any key to continue...'); pause disp(' '); disp('meanfunc = @meanConst; hyp.mean = 0;'); meanfunc = @meanConst; hyp.mean = 0; disp('covfunc = @covSEard; hyp.cov = log([1 1 1]);'); covfunc = @covSEard; hyp.cov = log([1 1 1]); disp('likfunc = @likErf;'); likfunc = @likErf; disp(' '); disp('hyp = minimize(hyp, @gp, -40, @infEP, meanfunc, covfunc, likfunc, x, y);'); hyp = minimize(hyp, @gp, -40, @infEP, meanfunc, covfunc, likfunc, x, y); disp('[a b c d lp] = gp(hyp, @infEP, meanfunc, covfunc, likfunc, x, y, t, ones(n, 1));'); [a b c d lp] = gp(hyp, @infEP, meanfunc, covfunc, likfunc, x, y, t, ones(n, 1)); disp(' '); figure set(gca, 'FontSize', 24); disp('plot(x1(1,:), x1(2,:), ''b+''); hold on;'); plot(x1(1,:), x1(2,:), 'b+', 'MarkerSize', 12); hold on; disp('plot(x2(1,:), x2(2,:), ''r+'')'); plot(x2(1,:), x2(2,:), 'r+', 'MarkerSize', 12) disp('contour(t1, t2, reshape(exp(lp), size(t1)), [0.1:0.1:0.9]);'); contour(t1, t2, reshape(exp(lp), size(t1)), [0.1:0.1:0.9]); [c h] = contour(t1, t2, reshape(exp(lp), size(t1)), [0.5 0.5]); set(h, 'LineWidth', 2) colorbar grid axis([-4 4 -4 4]) %print -depsc f6.eps disp(' '); gpml-matlab-v3.1-2010-09-27/doc/f3.gif000644 000765 000024 00000031525 11422240325 016625 0ustar00carlstaff000000 000000 GIF89aõš÷  """***---...///000111333555888:::;;;<<<>>>???@@@AAADDDHHHIIIJJJMMMNNNOOOPPPQQQRRRSSSUUU[[[\\\]]]^^^```aaabbbcccdddfffggnhhhiiijjjkkklllmmmnnnpppssstttuuuvvvwwwxxxzzz~~~tt†ee—ll“ÿÿ þ þ ÿýþþþýý--à!!û%%û ü""ü((û//ú22ù11ú44ù77ù==ø>>øccÕBB÷FF÷II÷KK÷RRõUUõ\\ôjjãïaaóccó``ôjjòkkòppñppòzzñƒƒƒ………†††‡‡‡ˆˆˆ‰‰‰ŠŠŠ“““–––———˜˜˜™™™žžžŸŸŸ£££¥¥¥  «¨¨¨©©©ªªª±±±³³³´´´ººº»»»¿¿¿ïŽŽíŒŒî——ìë¡¡ë««é¹¹çÀÀÀÁÁÁÃÃÃËËËÌÌÌÏÏÏÒÒÒÜÜÜÝÝÝßßßÆÆåÔÔãàààîîîûûûÿÿÿ!ù,õšþA H° Áƒ*\È°¡Ã‡#JœH±¢Å‹3jÜȱ£Ç CŠI²¤É“(Sª\ɲ¥Ë—0cÊœI³¦Í›8sêÜɳ§ÏŸ@ƒ J´¨Ñ£H“*]Ê´©Ó§P£JJµªÕ«X³jÝʵ«W¯/HK¶¬Ù³hÓª]˶­[·ÞÊK·®]º D¸Ë·¯ß¿fÅšÁO§Ãˆ+^̸±ãÇ#7¾âec5_¼øš°gŠ‡}žè¹à¡’S«^Ͳ™"E,/ÆLSóhƒ¥o? ­bî.µN¼8cNJŠ°iÜ óæÞCWÈ{:Céw~ßÎ5š"Q8þ1wn=†u†‡úœ_ø\ § ÝãËoüv#ÇÍk·ïíiýÂþþ!`twÌg`|vE^<–ßL¶(á„yÂn·ÜkE@â y†(¢A9ø‘I†(¶ÆrÊAF„û(£„™\˜âª±!FJˆçX&Ê!t•ÌJ‘2"ÉÂ$86ÙöEdLCiåt™œàä–±(†d/ ã•d~¶$—\nÂÆšlhQ„h°¹c™„Iå˜eæ¹U&&œˆæŸõaFjSÂT¥n™™¨Œ™ô`ØŸ.H…jvŠçgŽ˜iˆ5Bú'°¡‘ êiªV=ôá)—À¶¦jþ…¾tè©´Je¡Ÿ«6¹àÌ f©µ säê$¨Jx¨Hjœ8¬nH&)#|Æâè ¬ÅêÒ¬Âv‹ÔpT{#'Q08\¥²^êíºB‘ œ¸("+kèn«.»øò”%¼(*Û²Ã9ð‰~Ö-ʨˆ9Â/†$'*qÎ<ݦ!Rì ÈÉ®>¶FªÄù†ì Úe<ߎÉWoKÜŠìrM,¼krw qŸÊϾ¬³Lžx03‚®ng‚¶,ßëU´#"íþl' ›kÜ wHóÎX·„±ÓO—{EÇÃý(\õÙ'U‚×ÄmÌ£‡ÅÍÁ‚@VƒŒöÝ'Éöþpß‘`qpÌM·Ùœ<¢áë]°÷p5<Ü%1üV7ŒÖY<¡å½ Ròâª14q„\ HA“‹‰÷é%é͹dþñðps@qéw¢n{Hàºúc­¿ÞÚ 5 D»¥·Ï‘›ïÞX&½ë¾Õ Ÿ®uJ‡X=tª+ϘçEðJôÑnüøå®}c­gö¨—ˆ¾ôö’/ÿDšŸÏ‡_¬?*”0Ñó àC²g¿Nô úK¤#<ñyqœòâ ˆ˜ç˜#ÈÿXÒ2ÎdÐ?üŠ z@ÁNpbG®sÞjzp‡lp%  AA@å¹­Ž{\òB•Äp†þ¤–ý ± #äp81 DòÕoˆÉI™q&!¸%Ú ZFºÞhrÀ¤óµNŠÅ©ÒÔü°‰ãË€ ÷Ɖ¾};…`àû˜ˆÆÛebm»ÛÄ‚\—ÀÖÄ@‹á»bç‡G­îÉÁ¡|v‘2¢äŒ\ã„$É(C`äŽÆÖH:n%„ç¥V,´º6Âæò9%I¹A PÅZœCÕÇâ0²“‚t¥ñj8³FDñˆÝ¹C %âÈ“@R—h³Û@e„(ÀM>™¨‘(IçI­2@×ä %b°F~Ͳc¨¥q.AŠÓ$ÇD&Öà 0§ù6‰§q䀊”-—êþ<gV3g¦(ž1«™O—yBK3»¡ä¹I¸ "çĈ'z€/ØT:³²JÿtT+wˆeÆù¹ý`tæz[…©«Ê„‹¡Iä#í[%ëV®äõ6†Š% U×åê¤N+4p ƒBÁû-“Úšš¢o&"W5ÊH†%rˆ­nJ|Mp‡cÜ­þþ)jJ¸ÙªÈ©ã“T´:Kõ d"cmÄh"i&=*§Ø$•`éŠ5PQÐVy:‚µŽþ{¿¢Ì-µî ÏÌ^›¬Ò@À¢(Ïle—mMo¦0ŽlU|öÙ[FvRëÄ è‚“_nå¡C†å&5_ÔÈŽAc¨8KŸ±,Žz·°L”SÓ¥©‘BÌðZúM\7²étzÒ¡D.À‚ \`ÍEItŠH±9D6Ò¦ÃuM(q™Qõ¬Eðˆ1°Ú µöÈ­•­‘÷t11&#PÚœ¡¨]Æ«ª*ª ý•4OØ !!è$·¢¨íFænô–6KëTs;$=)càpá¡û@›H*MxÕßÿöˆ ú™Ü %§(ÊÂÛœvê~³;â$ydþ&Ññq£¨FèÓ,Ü‘lwdÛMù4t:ª‡pI&~JX3Ôºòò«Ǧ³¤ù³ê…ì31åÅ/PôдµÊÄNËqË!òŒü™1¾JŒvŒˆE ÿYÉ×=ôªçíÛŠÉ:¸ŽQϪJ˜×ÏÀå—sæfo)£vÄPÕ'Ňzê4UÖêTq7‘Â4Çô1Iì‰ l>Ÿ§?>cþˆÝ7‚÷¼ý1—ïÉ{²´wÇmÞc[! Í‘Ík¤óžGˆ«£&9iAN¯BŒáwBnšm|oÜÔ<âc^"&ô¢¯"Nö.ŸF¼ [{#4Ù“=sE„ù|‡ÌîqþṙEcà\! ~øwåÚ_±XãA?(ß& ›²ô½ªL ßüÄ¿H¥³kçdÿÛñt>73Ò7}µ“!è†|‡±u9±5ñU Â9ð®—°‡€ ðD!wËã6qoñÑ:qÆ5™ðl'tÔ׊ǂ$bzqSO6qpÅÁ" ˜1h€Ä£ávÛ‘ƒ.!mÜ1«3k8|‡*xÅA„ÊÓ„Tù¸„HÇ#¬‘E b+!r`å‚‹&“ŽVˆ’Ÿa·Ñ?LÇ ˜ˆa”Ù &™;0oÄá/=¢=LƒU9Wq^—™ÙÉ1Š!Mb©˜G)†)qš‘šÿ rn‘¡¸×°Ak"wý˜ š°ñlCa¹—j¹–ž`ŒQ.fÉZU™þ…ž˜‡<«ñTP§<' Ž&›¡›B¡’)¹I‹(Ì ZP‡ß¨—&ßÚÓ™çÉ—M!”ŸÔ0øÙ„Ɉúƒ–$qt«ñ°¹;öÙŸÑ)‘½Ç";)¸÷lðoÂŒ° h)F©d«“y*ž¡ž¸Ö{¢Y„Ù :É”6xÄ B’ÊSš)꟱÷‹±#ØC á˜᎑a¡ç¢¦é£K!™£ÁåhBo²‘9Ò ™ðtõ™‚‘•¿ÉŒÚ³@,¡¢Á¢‡Ö–à/“r—ˆQ.З c¹‹y>  ¸ÔUhk¾¸–½'¤Æá+þt¥£Šá¤ !‚¬AŸÚ3¬w›±U? ¥¦V¦¨ˆ‘&Úób”XÊ£"!Lá烤GèevQsÀžìñ§XÁª_Q™ ûȂʦj«—°±™lÓ‘.qN”ð©ªzPSz e¬a©€7#» *Ú…'€s0˜ž£a!¦™z«s D#¡!/ fDÕ§á wvQdfQH媿€ §ŒQ­ýqwÀ7º3º,~i‹Š”—ìz›†®b=H=°U{&eà“ÎcÉÉš*G°)TêFÊêö²Šbº;ü¹›æ ÉúþYÕ§å“dÀgkI/ ˆ[I­kÞv\ÊÀ Æ÷—©=ª²Á²%ðY0sPf2+A.v±€ЬÐ팀j”åÂëCÏ.‹±­ŽqœÑÂÁÌ?£Ó=ÑÓn|ÒQiò¼M=£†uщÑÒŸáÜ9ýk>AÕ ±Ð4ÒW©þ"yœªœ‚¹°Ÿ×Ï]jRÖ‘Êb¹føÓÄgχq×KÆqüœÎ7š*<²Á’ÑÑ3Ðn­°w††{=ÁÀ4©Ýö„{ÌëÜEûž)ŸÆR­;]/ÔɺìÉy»‡¡íyB˜Õ‡Ê-ð”‡ÕŠáÍp¼ÇaJš1„ Ð?QÛ·íÉþ Éì\0Òf<¼ØñЊñÂŽÌ Ô¼:ιèÜÏÝÙ#ÁÖ3ÑÐOq9{ù¸Ø6}©½f뜣3²CÅ„Ð$¡Þ3$låL”5„¾ß½Ú±/—<¶8F¡Ö Aà2„ÕËYÜR|#ä.µV +Ø…mÕ“WÀEpˆáÊQs‹Ç‘Î AÙÊw}½*ͽÙ0ár¦âѼÛU×þ¤Ü ÆŒõè؇¤ú1°'¿yß¹¢¯Kaw½•ªç ¶(‹á­ÁŒñ¢çp‡€Ü!š0,~`ÞGaw„­ÛÖ •Æà‰aæíØtn‰Eð¡?Лâ ÍÆÑÜü'ÃÞ¬,eä6…M7ºÅÅ]“Á'*DÜtü~$Ðç7½áÆRoN‘màd_ÞËq•›Cû­ŒÓßM; •? èïx‘Õ2”n,Þâeìo<ÚÀÈŠÑä Ñ ©8„p1ð…«aå&ïQQï·sðzóÎm•–ÝŠ!åAåP/ÝsíOïPªtm¶þÒçñ„žêî­Á¤«ÂðRaááòãÞb®`£é!ÁK…Ž!t%Æ\öñ7Qž0½bs 4O>Žâ–~/q$´­SêŽÙÀÎìÙa¦Ë°ÙK?>{\@ßÍZÿmˆõòAêžâÌUQLžÐ¿‚О 6ì‡aßï<më‡q“Eualõvúë8âèoÿ‹`Ñ´(_ÕÇ~ÙØ P¥v7`ž#!Þ^ÍèÜÁö¢ø‹ª„|}…ïk)lhœÝ)È}|qN›˜9í#ìCÞ±ê&ýÆ•ötŠaÔ‘•r¹ö'‹&0òSû¡ûÓ(ô}þVi¯!®zýÈ›ß!ý²`'PýOQLwf1Z%éyN=”£ÀÜﶯ„¯oB O©ú\"ÃZQL†'CEݹéüDV²pàÑI @  D˜PáB† I „(pL/#²±˜±S”"5~r  MžD™RåJ†Xj A'PžÁ,ÁÒä‹:}þTèP¢Eƒ²¸4pS‘"›f"”è΄4>2RäÑG!ÙQÂIäX²2MÊ@sêZ¶ ]ªÔÀ\ºuíåÙVï^¾}ý®QQ‘(ú»RŒ^Š\á*r“’Že-‡Ì)ñfŸoSÚþoOÎ¥‰j6h¦¿¢˜ „jƒ>BbjFc×Æ„/ÿÎCmâ =G˜ùr†j™??xçÄ+EEœM;G!®`ÅZäq°VàåÞ‘ ]õqæÊÕ¿‡¿i&œ˜B‚è§ñJ?>râ66$P tj N(ƒÍ<àЊÏ4öLŠ©!˜Šr/B 7ôë¡+¢*).³aj«‹@BÃ;ËŒÃÄ&l¨B† CÒdä±Ç¡29Á5Šb+¨8&É1ÈV¼Š©F^üM8û¢¡CBç,D.ˆª¨ " Ä¿–*B7×4‚„²Š ´þÌ"¿l«Êƒ®ÄÒ®ôØOA=¨$J¢;àXâ-<â|ì;:Ǻäç-JO…lì+CNGåÑÄN8±½4òÐZ,¢ÁN$¥4¬KËŠ1R;]IƒýüuWaß[ ¢D‹(æù¾ÿ`mP‘ˆjÅïÖ±ô–(OU 6Û PóV¨¸h"Þ.!Üâ< V1ªh6(S^kAº$­pƒÚ¡>ë t_¡6X§ "r¢ˆ'Z”`íÈm—)ÂØ0Ã1¦ê½w¬\ þ©ßƒþéL¢ºå¸ä½2‰Aa]ã@"XE®Ø±©Nb¦®PÄÒŒ5ÊÄôLþ^Éc“*éÔ¡H:é¢à€¸“Á”ÀN=GËÚD1¼@ÃÞ?ÊÔa¥[*J~=zǶÉᘿö`ˆZ´ÂÙN.) :¦·Æ[J¶Q% x*AX¸ B ü˜rí½ÊÄ„ˆªûBÑ;@qœ6 ñÞ¹;o¨ï“.ÙìÓ2˜#)‹b`ÁkÐ}„c:™"¢ÔenjέÕôu“"0*€KŸê š¶È(ñÏt!BQ Þ>.öÝoýÑæÁÊ“œø¥OP_.ß¾GO‚R ç®[½õ¯‡ò,êÓ?¨oNü£Õ$Ë Ð§!ܽN²Huþ®Ã0@¢€F*ý€#›L<0}úëæ “ /!H[IÄt 0>®Û .!7ûÁ^·3 UsÆERô»Ã~dè;ÑMÅð`rLçÄ•§„øÓ'½A$;ñQ¡ %‘|í°y=,Ê!ä²¢D’3Ï‘8Œp _ãJqŠÙØ‚E¡|` é¢I(qç ”ãsÜF6YnCèãnÞØ B<1ù+z0ɹpqˆ*ñ"étæ½'sl»Ã"Ò‚ë“ÅÑC³DrCÎ\ d3ѹ€¯Y>©$/™ Ï<™‚å#iþ£I&B͉2$HX¹µ9hO˜\‰„š?é$— !&Ùƒz”ñ&(àйA©ká4«Ð‰9p~3†î¬ <שNyÖó ô¼g;Ëyê3þdç;ó)Ðz¨aû¨Ø|4ž}¨CûiІ´&e´ˆÊd ½/cù2?í)Ñ‘Zt¢…H!zR––T“Øæ —’JÌåºÅ!ìL”\Sˆ>¡Ä­U}=Ó4Ä$¤›ÎcB %S‘Á)åØ·`ƒ ™r²Ÿ$ ñI'¢Zq3LÚæFyÛ© ­È ¥W$ˆ…¼•‡+ €–‚þ€°¤«ù*BX༢¨È1«P*‡þÅçH‰^Ô’XÕdÀ*‘¹^BY¶éOd !Jþ ŠÀd ÝQíJ(´e ~øH‡Õ–F‰A£}” >bvwq4*Bú–׆ðÕ¯Aôê.â Ðoê XW ¤gçƒNkÇ(µÒ™Ò!؇ÛÝe¢d .4SRÕ«&$«[-r«\…€åÄ0‡•I§uË©„+3’¨œ ¶ê!~ãu=pö‚+©)nº';F7!£ ¬ P›YìÒÆÂ,É„eEâN'“cMUl”•žBÑ—ôÓÛB>ü8þÑI34Õ<®%S’ÉEV0ôíÑ«û‘Lè—6=èÃnŠÆLb>Êä‹ñÖ³Ÿ¥W¸:™%ŽærâR¾¤²A2 åëÍVoÇʵ§Fà ˆ!2@rÉ%ˆÁš?zX+«÷9y„ÆiX43Ç2¦ÚÌk„M È A,Î÷A‹ûLÇPY6‹´H&ÊZšŸd°å1òfìû‘Áæ˜u’'zp‚íîl–ÎlH©;ûž.’ Ñ f$Ÿý2i„À€ÀIµ_X ʈ¡Lér¿”‰\z6çv,LNó„¶FFÍ#ñ¾¨È‹æ ̨'þA„±Ùº ŽmÎ è:–¼V.oÃý¢òHÍtJv[Ô©ê)& ö†î?\ú2QísèðÝ1ïû#9ä'.ðë±@z ŸÊ²W¹äˆÜOXžÈ°=–?F\âê!B-à”/3^È©Ë#©skZ±³ Gmô"ç¢Dúk”1õ$GÿZ·÷Ò-ÏœÈ]‹±ÄÊLb*=Ç#òhw7V`²ôXæPt—gâ ©¸Õ?‚-6 uç4g»O*¡ô²¬Á7‘4Ð*}Y²°®í&‘º^‚¥u¹ƒ¸ÏÑøÊW’9ü÷'‡`ù*H.<› JËdžE2‰þž{ÐU€Þø¹õîd:94áͳ̑\:9¨ûC»Ê ±·‘u[d®\ ·ß~¤²^#>ã g]ˆÕ_å@là(<ÚdÀæ ÉH úß  ™ãšS0°ÏbPùAá¦|˜›-mÎÉ  ®—é_Šö™„ÄÜÉ„CØñ#³±ø .ˆˆÀ{›È ¸€ø>Ó{9ÔÛª>üCÄŠ;:ù?‹¸„ˆ T¸€€ƒZs*(‚5  4½Äk å81ü‰¦#ãIÀßAú!2` 082<<d åè%瓲¶¢ˆ.O˜þ.É»Œ œ¢¯¸’³SŸ ®ýK&\‹¼!üÓÂN ;ÎÀƒÅÙd=ÊXr@%Œ83œŠÈ0r¾5¤ Í8„ñÃF`½‚sr$<´2=4Š8ÞRC˘“c‹Ü#?̘qŸ÷™„ßcÄàrD ‰Ùc½5„£  u¡„ñÃBM,ÊP„GP‘Lø0„µÃ ¾2|èXDŹ™©¢P OØ»Nð¸Ž{6˜E9Dï€xJ¸×¹Æ‚E¢Ð9«Ó¨¹¾í 5Œ[ 8|*‘P/  ›a /ˆÂÆkY±® ÅM{nœÄß@¾¢±âþó­xŽ%ð‹‘*È«k¡FЙBÅ{|¦m<t´8T9£YD ¸ƒ°+ ÚÛ„dƒGÐT°ð.` '°ˆ[”H|T} ·M0g 4P9¼ ”QÅ•¸³Û±Œe%[Ʊ›¦h|#g£G¨F”C£"C ¡H¡ÉEâ„tÇXQJ‘ÀºŸØ»$/0Ä0ƒ›Gà„¯ ­‘ †´ˆ xI˜D=‹œ"H È(ÉNh„4hÇ+èÊ GÜ“ÄøÆÉ>ÙN€–À<‘vdƒÕ‹Ç[ù W{Š¨˜ËFÌG»¤jÁ-zqK‘£•¸ÍŽ®ÀþÆdŠ¦¤ššaÍ7¹Ê”ÃÈÌôªÊ£àÌÝQʈ‚ÐœzYËÈ´•àíËÈšIÉ߀/Ì`TG‘,”¸=]¤?è¸ÊëÀ ‰M¨™ÉT4 “ €ÌÂ3‚ï,ììJç Ž¸!{<‰l|Àê|ŽëäœÉ Ç„N¨Œ)‹"›B„X¸dt£` ó4¿kKð°–ñ(J?ÀµÚ”ÏöÈÍ­qŒ(ÀÏQª ½ ÕZ.B|‘fä(0`Š- ½ŒGØM°Ðõüˆ Ž¨„Ð‹É Í¶LK ”9±ŒB8h¬Á A(‘–TyM¡EHH$ ƒ Gó(¸¢”þÅͼKHŽ‚sÎ"û ¸˜ƒIèÀË $à„ =Ñ­ðL`ÑŒpŒ×ü@•€QašJ“¹M¼ Ñ[ÑOdÙ¹v+0-긂%±Z‹4]ÓˆH•¢ü.§“Ò1ªÓŸ ÏUQq6[ùÃûÐ<1ƒ!LŒ9Tˆp ˆØÅFåÅíÌÜx!¹Æ{„‰(.`ƒ'¸>µ ZÌÏýl„Pˆ³$ŒØHTE¢Gõ‰¸S:U:)¸àœ¹!96ƒŒÑÕú(Óš©œ¡®{²“8ÕæùÖ’1VˆÔ[I•ýJå?åÇ°•¡Ö÷ÑÕ<-\% @ Ì9`Ô†þ€Ï½é×pW–(×Ká&£h-Ò7:–`…×{©ÖN¨VN°É5eKßì„;$Ö©¢RÎaPk)8t]ÕgkØA•WZN3µÜð<‹°ÄŒÝž€]‰mUÙ̘ EÙÉSA_µˆ‡}X5eŠÞ´Œœ‰OtÙG‚Y•Ù­ƒNk±Ož=Í"@‚Y¤ÅXe„ª%Éx…,zIN8á%PÔè\<‰8}¤9W^KÖò°Ï®¥“…µX¼)¸4ÀÚX­Û› yÙÔN=x©/Y§ÎÖ£}ÙÝ™±ÌQ¬¤ ƒÍÈÝùÙüdƒ¬ J€•+[ˆhYÃ}¤M‰¥ŽMÈ þ{…•u[ mÚÈåœÉ;„¼g¤E6@H40Sõ\UÕ©Ìëù °• `MK¨ ×íÙšÙJ°XH‘€!,3½påÐE dÅYY¼W…\¼ÙÖtÄã͈GH4ØO/Ø˲(Ü÷4½õé5µµÍ¿»žTqÖ­Ô€ÜAëÂ_‘ðÞŒq²ÜÕݺÌh%^‘PWΩØò˜88 ¨ èèȸƒdÜß{y7ÿ-VÄ¥ÌÕÝhÍÞAeÕËP`È™Ð#ô8vÝš§ä`üa_“]Ë Ù7jÖê’:Á€p0Y#>‹;ߧ3=´^¦“§ÁÛ[)à{iþâkŠ;8`ÃÝ–a$†ö6¾¶œ_€Ëë¾[¤Áâ,ÞÝK%ΆˆT:(ÁÒ²¨<¶Ø´ãœÎMã¨S[kiS(ñ ŽZà=]9)‹1Ö‹Ž|EsêÖ•€^ЉdÝLxá €ÀÖ_¦ØÜ[`Ú¹ËpUít<Õбж½È=8. þ õ½õ‹JÐ ˜ádPMR½ ’P—âÓÇ}‘JÛËüÏ©¸?l%YîãAIZ úƒè ½,÷ÝN>…‡)Íß1\ž)ϲh°¿èfË/>ŽæTM ±qOP€ @êåîX\þ~ЃЃñúH¦ÐIË(g²ø}m‹Ï[‰v>ÜÍøž:;( ;ˆÁre‹€VÓåšaTˆ4¼Œ4`+€.в¸„!n ûšè±`g"><#ÖFÎÀ‘µ©©™¢^!a\éIæPŠ–‚&‹€.¼šæ¿Â´–þUè+âŒÑº&Z£.~xV ³ G ®ÌÇ|„îÜÑGö‹rËÚj¶ñ„CëCžÍ0jBêA#ødèÔh)Æ+^¬M7Ñ"‚0®ÊÕé²hkŠ¨ŠR)’Šhu¾Ϩ¨ÃþëÄöëÅF)Äfì–Z)—RìÆžlÈìË6)É~lͦ쪄°¾ƒþ±NŒ²ŽhmÖˆ@Ž6D ŒëªM„DxEœ´ðmG/Xk‹ +Úظ’Έ<ç¥e¾ ílþÐûpc0éL L°;ýU¼´/ È‹±èŒXæÄз[AãîIßáÞ ù¾—.íi`¶Œ“.¶_$‹‡å„F@%°Üwdƒäî8ph7ÓcàPï®>b†‡›©gˆ`¼‹„Ö  ŽkŠ†gLf‘€áâ(±‹Eño¢Þ ±ù,1 ðòž=øåG4£gnð­QmâÈA’IÎpB™æ±±fÀfP(ðú°^dëï„Ю±È`æF ü%¬Ö»"ï½°þeÏeØeÁ2m!1&àhA «±(Ó­êç $3Ç{ñ½©äKvòæõª(¥¹ ?ØoúYnùi€FŽˆ7 sЙᆸñ6F`#W _ 7;‡0¿s?^ã˨Ž„ ®þ§>cŸ&Žˆò^ gL__-öqO6‘V ¸~#Œ†ŽLºéØ0ôù;õ€gÇs`_“½þ £]‹'¼ž@”‘9¢PéÇi÷}qu.2íÞþíÓ±e‹FÛ-Ï8 ì:G7oßµD‰ŠvPÿ ?àç­Iu éH¤|!v€‡wpÇ­&2g‰[ç/ñAˆ ªˆ—x€ýã²PsàÀð½X5f"è ±ÂpùÀضÐt¥™ùa‰w…¨a%vS¾æ³–j—#V7¡'•›Oˆ.ù߈ҾªIñ‡zP0úäX[7}ß‹"²Ù ú¨ÏØ©q‘Húߘ²¸¥{’ðúµ—zŠ«ïÓ ÷‹ òš˜wg¼ßt׈±çGFcðòxz¶x°7™…ûò(„¼ã‹¦‡¹'|‰7|¿S¿¿Œù°1ñßÈöþȇúɇuHüòú¾Hgóà­®»7Ö/zŠ`G¦£rÆ&Õ?»$Ÿý ý]ûŽˆ;ÈoΠ{)ôüµŸüH}(10Ï€S‡ˆ.~¯}ÎÌÅ¥“CŽ]/ µþ¨?~Îd“뽈4=ÂÂIÜß—Ìǽϖé×gÓY){æ0ÿÕË×ï~Týþïí½ˆž@,hð „ ¬P§ˆ'‚8Äð"ÆŒ7rìèñ#È"G’,iò¤B (K²ð3Q¢—"^^Ò¬i³‰•#=](d3ÆB‡-jô(Ò¤GU*ÍØ’&£"E ݬZ³PŽ¦=‘8þqGâ¥9à̤õlQ³h×¢TËö-F¦p >•È‰“’"h¬òøCÐÜ„™àd¨ƒDЂ31ä‹#–˸®D4E”pêë僙(….mú4êÔ-~áRâ#©<Vmû6îܺO²ž‹¹S™´ùf"”u7òäÊ—Ûî 3›Í›†óýaQw%æl³k?˽ûJço]GÜ$• u¾À³oïþ=IñlëƼ’Þj¦œð÷óïÏ_þZOE7Õ}UùA– *¸ n¢Õ$çXU ¤íæƒ&a˜áHrȃg±ðGpöMhS&8$7Ù‡"±ØâG/ˆPˆZ½Ð…tþ'ÚtÉ$3úø#(ÕØÔR)¢£M~|$“M:WdP!’6éñ$–Y>9¤R0tÑY•4í¡ÜwZ.d¡™ •é$—IµðZ˜]BHšuÚ¹`›Hµ”Iœé±ä Ú^žG‘קDs º(£ÊjÔo}þ°œ‡Vjç¥@>ZT¤aNbÇr2Þ)j¤þ¸)QV G¦ºú*d¨u(¢5Àz+®‘É*”ªHª˜+°Á¢µ«N½ž8ÉÌ%¶hq.»edÆNšÉõÐè!×-zm–Ä®Dk˜™€Æ·‹fÛh¹ÐB&mã.—® ç2 o“Þ¢Änz…л[ ØöÑh¿ÝFÖþœaþÀ$\à ;¬0ÃKœðÃ+lqÃC\ñÆOÌðñÃ$k¬1Ç(gÌ1Å—Œ±Å-‹üòÆ3«¼°ÉÛü1Ë:ãLóÏ'¬rÎ) s͸LtÈ1û\ôÊPKM4ÑKóµÌA_ôÅ“ û5Øa‹=6Ùe›}6Úi«½6Ûm»ý6ÜqË=7ÝuÛ}7Þyëö šÉ À ­bùƒß{Û‰@âwV"80Àšfz¸âiVR/€×  ŽÀ‡hÐyæYêÑ9ë®®ÁA§_iæ ¬£žú“‡ô°ûívÖ>{A«¿¦»÷Ð;îX"Ÿ&¬Yþ º|òMR¯åõ×g©}õ?rÿdöž§ù}÷0’o½øÏäúå+§üñËBíþôé»Ïžíý·Žý~Ô¼5"zùÛúõ»ƒï€ R}ôºØxuŠà÷ƒÁ1®\Ç#3è£ ÂèqˆÜä*g&Šú3àG8Ãï…-¼!s¨Ãò°‡>ü!ƒ(Ä!±ˆF3, if norm(s2)>0, s2zero = 0; end, end % s2==0 ? if s2zero % log probability evaluation lp = lZ - 2*logcosh(tau*(y-mu)); s2 = 0; else % prediction lp = likSech2(hyp, y, mu, s2, 'infEP'); end ymu = {}; ys2 = {}; if nargout>1 ymu = mu; % first y moment if nargout>2 ys2 = s2 + sn.^2; % second y moment end end varargout = {lp,ymu,ys2}; else % inference mode switch inf case 'infLaplace' r = y-mu; [g,dg,d2g,d3g] = logcosh(tau.*r); % precompute derivatives if nargin<6 % no derivative mode dlp = {}; d2lp = {}; d3lp = {}; lp = lZ - 2*g; if nargout>1 % first derivatives dlp = 2*tau.*dg; if nargout>2 % 2nd derivative of log likelihood d2lp = -2*tau.^2.*d2g; if nargout>3 % 3rd derivative of log likelihood d3lp = 2*tau.^3.*d3g; end end end varargout = {sum(lp),dlp,d2lp,d3lp}; else % derivative mode lp_dhyp = 2*tau.*r.*dg - 1; % derivative w.r.t. sn d2lp_dhyp = 2*tau.^2.*(2*d2g + tau.*r.*d3g); varargout = {lp_dhyp,d2lp_dhyp}; end case 'infEP' n = max([length(y),length(mu),length(s2),length(sn)]); on = ones(n,1); y = y.*on; mu = mu.*on; s2 = s2.*on; sn = sn.*on; % vectors only fac = 1e3; % factor between the widths of the two distributions ... % ... from when one considered a delta peak, we use 3 orders of magnitude idlik = fac*sn1/(2*tau^2) & ~idup; % interesting zone is in between [g,dg,d2g] = inv_xcothx(2*tau^2*ga(id)); thg = tanh(g); h(id) = 4*logcosh(g) -2*g.*thg; h = h + y.^2./ga - 2*lZ; g1mt2 = g.*(1-thg.^2); % first derivatives dh(idup) = 4*tau^2; dh(id) = 4*tau^2*dg.*( thg - g1mt2 ); dh = dh - (y./ga).^2; d2h(id) = 8*tau^4*( g1mt2.*(2*dg.^2.*thg-d2g) +d2g.*thg ); % second derivs d2h = d2h + 2*y.^2./ga.^3; id = ga<0; h(id) = Inf; dh(id) = 0; d2h(id) = 0; % neg. var. treatment varargout = {h,b,dh,db,d2h,d2b}; else ga = s2; n = numel(ga); dhhyp = zeros(n,1); idup = 50<=ga(:); % asymptotic behavior for large gammas dhhyp(idup) = -2*pi^2/3/sn^2*ga(idup); % tau = pi/(2*sqrt(3)*sn) id = ga(:)>1/(2*tau^2) & ~idup; % interesting zone is in between [g,dg] = inv_xcothx(2*tau^2*ga(id)); thg = tanh(g); [f,df] = logcosh(g); % h = 4*f -2*g.*tanh(g) if any(id) dhhyp(id) = 8*tau^2*ga(id).*dg.*( thg - 2*df + g.*(1-thg.^2) ); end dhhyp = dhhyp + 2; % from lZ = log(pi) - log(sn) - log(4*sqrt(3)) dhhyp(ga<0) = 0; % negative variances get a special treatment varargout = {dhhyp}; % deriv. wrt hyp.lik end end end % Invert f(x) = x.*coth(x) = y, return the positive value % uses Newton's method to minimize (y-f(x))² w.r.t. x function [x,dx,d2x] = inv_xcothx(y) % f = @(x) x.*coth(x); % df = @(x) x + (1-f(x)).*coth(x); % d2f = @(x) 2*(1-f(x)).*(1-coth(x).^2); if numel(y)==0, x=[]; dx=[]; d2x=[]; return, end x = sqrt(y.^2-1); i = 1; % init ep = eps./(1+abs(x)); th=tanh(x); fx=(x+ep)./(th+ep); % function r = fx-y; % init while i==1 || (i<10 && max(abs(r))>1e-12) dfx = x + (1-fx)./(th+ep); % first derivative d2fx = 2*( 1-fx + (x-th+ep/3)./(th.*th.*th+ep) ); % second derivative x = x - r.*dfx./(dfx.^2+r.*d2fx+ep); % Newton step ep = eps./(1+abs(x)); th=tanh(x); fx=(x+ep)./(th+ep); r = fx-y; i = i+1; end % first derivative dx/dy; derivatives of inverse functions are reciprocals dx = 1./dfx; % second derivative d2x/dy2; quotient rule and chaine rule d2x = -d2fx.*dx./dfx./dfx; % numerically safe version of log(cosh(x)) = log(exp(x)+exp(-x))-log(2) function [f,df,d2f,d3f] = logcosh(x) a = exp(-2*abs(x)); % always between 0 and 1 and therefore safe to evaluate f = abs(x) + log(1+a) - log(2); df = sign(x).*( 1 - 2*a./(1+a) ); d2f = 4*a./(1+a).^2; d3f = -8*sign(x).*a.*(1-a)./(1+a).^3; % computes y = log( exp(A)*x ) in a numerically safe way by subtracting the % maximal value in each row to avoid cancelation after taking the exp function y = log_expA_x(A,x) N = size(A,2); maxA = max(A,[],2); % number of columns, max over columns y = log(exp(A-maxA*ones(1,N))*x) + maxA; % exp(A) = exp(A-max(A))*exp(max(A)) % computes y = ( (exp(A).*B)*z ) ./ ( exp(A)*x ) in a numerically safe way % The function is not general in the sense that it yields correct values for % all types of inputs. We assume that the values are close together. function y = expABz_expAx(A,x,B,z) N = size(A,2); maxA = max(A,[],2); % number of columns, max over columns A = A-maxA*ones(1,N); % subtract maximum value y = ( (exp(A).*B)*z ) ./ ( exp(A)*x );gpml-matlab-v3.1-2010-09-27/lik/likT.m000644 000765 000024 00000021122 11422240322 016706 0ustar00carlstaff000000 000000 function [varargout] = likT(hyp, y, mu, s2, inf, i) % likT - Student's t likelihood function for regression. % The expression for the likelihood is % likT(t) = Z * ( 1 + (t-y)^2/(nu*sn^2) ).^(-(nu+1)/2), % where Z = gamma((nu+1)/2) / (gamma(nu/2)*sqrt(nu*pi)*sn) % and y is the mean (for nu>1) and nu*sn^2/(nu-2) is the variance (for nu>2). % % The hyperparameters are: % % hyp = [ log(nu-1) % log(sn) ] % % Note that the parametrisation guarantees nu>1, thus the mean always exists. % % Several modes are provided, for computing likelihoods, derivatives and moments % respectively, see likelihoods.m for the details. In general, care is taken % to avoid numerical issues when the arguments are extreme. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-07-21. % % See also likFunctions.m. if nargin<2, varargout = {'2'}; return; end % report number of hyperparameters numin = 1; % minimum value of nu nu = exp(hyp(1))+numin; sn2 = exp(2*hyp(2)); % extract hyperparameters lZ = loggamma(nu/2+1/2) - loggamma(nu/2) - log(nu*pi*sn2)/2; if nargin<5 % prediction mode if inf is not present if numel(y)==0, y = zeros(size(mu)); end s2zero = 1; if nargin>3, if norm(s2)>0, s2zero = 0; end, end % s2==0 ? if s2zero % log probability evaluation lp = lZ - (nu+1)*log( 1+(y-mu).^2./(nu.*sn2) )/2; else % prediction lp = likT(hyp, y, mu, s2, 'infEP'); end ymu = {}; ys2 = {}; if nargout>1 ymu = mu; % first y moment; for nu<=1 we this is the mode if nargout>2 if nu<=2 ys2 = Inf(size(s2)); else ys2 = s2 + nu*sn2/(nu-2); % second y moment end end end varargout = {lp,ymu,ys2}; else switch inf case 'infLaplace' r = y-mu; r2 = r.^2; if nargin<6 % no derivative mode dlp = {}; d2lp = {}; d3lp = {}; lp = lZ - (nu+1)*log( 1+r2./(nu.*sn2) )/2; if nargout>1 a = r2+nu*sn2; dlp = (nu+1)*r./a; % dlp, derivative of log likelihood if nargout>2 % d2lp, 2nd derivative of log likelihood d2lp = (nu+1)*(r2-nu*sn2)./a.^2; if nargout>3 % d3lp, 3rd derivative of log likelihood d3lp = (nu+1)*2*r.*(r2-3*nu*sn2)./a.^3; end end end varargout = {sum(lp),dlp,d2lp,d3lp}; else % derivative mode a3 = (r2+nu*sn2).^3; if i==1 % derivative w.r.t. nu lp_dhyp = nu*( dloggamma(nu/2+1/2)-dloggamma(nu/2) )/2 - 1/2 ... -nu*log(1+r2/(nu*sn2))/2 +(nu/2+1/2)*r2./(nu*sn2+r2); lp_dhyp = (1-numin/nu)*lp_dhyp; % correct for lower bound on nu d2lp_dhyp = nu*( r2.*(r2-3*sn2*(1+nu)) + nu*sn2^2 )./a3; d2lp_dhyp = (1-numin/nu)*d2lp_dhyp; % correct for lower bound on nu else % derivative w.r.t. sn lp_dhyp = (nu+1)*r2./(r2+nu*sn2) - 1; d2lp_dhyp = (nu+1)*2*nu*sn2*(nu*sn2-3*r2)./a3; end varargout = {lp_dhyp,d2lp_dhyp}; end case 'infEP' if nargout>1 error('infEP not supported since likT is not log-concave') end n = max([length(y),length(mu),length(s2)]); on = ones(n,1); y = y(:).*on; mu = mu(:).*on; sig = sqrt(s2(:)).*on; % vectors only % since we are not aware of an analytical expression of the integral, % we use Gaussian-Hermite quadrature N = 20; [t,w] = gauher(N); oN = ones(1,N); lZ = likT(hyp, y*oN, sig*t'+mu*oN, []); lZ = log_expA_x(lZ,w); % log( exp(lZ)*w ) varargout = {lZ}; case 'infVB' if nargin<6 % variational lower site bound % t(s) \propto (1+(s-y)^2/(nu*s2))^(-nu/2+1/2) % the bound has the form: b*s - s.^2/(2*ga) - h(ga)/2 with b=y/ga!! ga = s2; n = numel(ga); b = y./ga; y = y.*ones(n,1); db = -y./ga.^2; d2b = 2*y./ga.^3; id = ga<=sn2*nu/(nu+1); h = (nu+1)*( log(ga*(1+1/nu)/sn2) - 1 ) + (nu*sn2+y.^2)./ga; h(id) = y(id).^2./ga(id); h = h - 2*lZ; dh = (nu+1)./ga - (nu*sn2+y.^2)./ga.^2; dh(id) = -y(id).^2./ga(id).^2; d2h = -(nu+1)./ga.^2 + 2*(nu*sn2+y.^2)./ga.^3; d2h(id) = 2*y(id).^2./ga(id).^3; id = ga<0; h(id) = Inf; dh(id) = 0; d2h(id) = 0; % neg. var. treatment varargout = {h,b,dh,db,d2h,d2b}; else ga = s2; n = numel(ga); dhhyp = zeros(n,1); id = ga>sn2*nu/(nu+1); % dhhyp(~id) = 0 if i==1 % log(nu) % h = (nu+1)*log(1+1/nu) - nu + nu*sn2./ga; dhhyp(id) = nu*log(ga(id)*(1+1/nu)/sn2) - 1 - nu + nu*sn2./ga(id); % lZ = loggamma(nu/2+1/2) - loggamma(nu/2) - log(nu)/2 dhhyp = dhhyp - nu*dloggamma(nu/2+1/2) + nu*dloggamma(nu/2) + 1; % -2*lZ dhhyp = (1-numin/nu)*dhhyp; % correct for lower bound on nu else % log(sn) % h = (nu+1)*log(1/sn2) + nu*sn2./ga; dhhyp(id) = -2*(nu+1) + 2*nu*sn2./ga(id); % lZ = - log(sn2)/2 dhhyp = dhhyp + 2; % -2*lZ end dhhyp(ga<0) = 0; % negative variances get a special treatment varargout = {dhhyp}; % deriv. wrt hyp.lik end end end % Returns the log of the gamma function. Source: Pike, M.C., and % I.D. Hill, Algorithm 291, Communications of the ACM, 9,9:p.684 (Sept, 1966). % Accuracy to 10 decimal places. Uses Sterling's formula. % Derivative from Abromowitz and Stegun, formula 6.3.18 with recurrence 6.3.5. function f = loggamma(x) x = x+6; f = 1./(x.*x); f = (((-0.000595238095238*f+0.000793650793651).*f-1/360).*f+1/12)./x; f = (x-0.5).*log(x)-x+0.918938533204673+f; f = f-log(x-1)-log(x-2)-log(x-3)-log(x-4)-log(x-5)-log(x-6); function df = dloggamma(x) x = x+6; df = 1./(x.*x); df = (((df/240-0.003968253986254).*df+1/120).*df-1/120).*df; df = df+log(x)-0.5./x-1./(x-1)-1./(x-2)-1./(x-3)-1./(x-4)-1./(x-5)-1./(x-6); % computes y = log( exp(A)*x ) in a numerically safe way by subtracting the % maximal value in each row to avoid cancelation after taking the exp function y = log_expA_x(A,x) N = size(A,2); maxA = max(A,[],2); % number of columns, max over columns y = log(exp(A-maxA*ones(1,N))*x) + maxA; % exp(A) = exp(A-max(A))*exp(max(A)) % compute abscissas and weight factors for Gaussian-Hermite quadrature % % CALL: [x,w]=gauher(N) % % x = base points (abscissas) % w = weight factors % N = number of base points (abscissas) (integrates a (2N-1)th order % polynomial exactly) % % p(x)=exp(-x^2/2)/sqrt(2*pi), a =-Inf, b = Inf % % The Gaussian Quadrature integrates a (2n-1)th order % polynomial exactly and the integral is of the form % b N % Int ( p(x)* F(x) ) dx = Sum ( w_j* F( x_j ) ) % a j=1 % % this procedure uses the coefficients a(j), b(j) of the % recurrence relation % % b p (x) = (x - a ) p (x) - b p (x) % j j j j-1 j-1 j-2 % % for the various classical (normalized) orthogonal polynomials, % and the zero-th moment % % 1 = integral w(x) dx % % of the given polynomial's weight function w(x). Since the % polynomials are orthonormalized, the tridiagonal matrix is % guaranteed to be symmetric. function [x,w]=gauher(N) if N==20 % return precalculated values x=[ -7.619048541679757;-6.510590157013656;-5.578738805893203; -4.734581334046057;-3.943967350657318;-3.18901481655339 ; -2.458663611172367;-1.745247320814127;-1.042945348802751; -0.346964157081356; 0.346964157081356; 1.042945348802751; 1.745247320814127; 2.458663611172367; 3.18901481655339 ; 3.943967350657316; 4.734581334046057; 5.578738805893202; 6.510590157013653; 7.619048541679757]; w=[ 0.000000000000126; 0.000000000248206; 0.000000061274903; 0.00000440212109 ; 0.000128826279962; 0.00183010313108 ; 0.013997837447101; 0.061506372063977; 0.161739333984 ; 0.260793063449555; 0.260793063449555; 0.161739333984 ; 0.061506372063977; 0.013997837447101; 0.00183010313108 ; 0.000128826279962; 0.00000440212109 ; 0.000000061274903; 0.000000000248206; 0.000000000000126 ]; else b = sqrt( (1:N-1)/2 )'; [V,D] = eig( diag(b,1) + diag(b,-1) ); w = V(1,:)'.^2; x = sqrt(2)*diag(D); end gpml-matlab-v3.1-2010-09-27/lik/likLaplace.m000644 000765 000024 00000025477 11474502346 020103 0ustar00carlstaff000000 000000 function [varargout] = likLaplace(hyp, y, mu, s2, inf, i) % likLaplace - Laplacian likelihood function for regression. % The expression for the likelihood is % likLaplace(t) = exp(-|t-y|/b)/(2*b) with b = sn/sqrt(2), % where y is the mean and sn^2 is the variance. % % The hyperparameters are: % % hyp = [ log(sn) ] % % Several modes are provided, for computing likelihoods, derivatives and moments % respectively, see likelihoods.m for the details. In general, care is taken % to avoid numerical issues when the arguments are extreme. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-07-21. % % See also likFunctions.m. if nargin<2, varargout = {'1'}; return; end % report number of hyperparameters sn = exp(hyp); b = sn/sqrt(2); if nargin<5 % prediction mode if inf is not present if numel(y)==0, y = zeros(size(mu)); end s2zero = 1; if nargin>3, if norm(s2)>0, s2zero = 0; end, end % s2==0 ? if s2zero % log probability evaluation lp = -abs(y-mu)./b -log(2*b); s2 = 0; else % prediction lp = likLaplace(hyp, y, mu, s2, 'infEP'); end ymu = {}; ys2 = {}; if nargout>1 ymu = mu; % first y moment if nargout>2 ys2 = s2 + sn.^2; % second y moment end end varargout = {lp,ymu,ys2}; else % inference mode switch inf case 'infLaplace' if nargin<6 % no derivative mode if numel(y)==0, y=0; end ymmu = y-mu; dlp = {}; d2lp = {}; d3lp = {}; lp = -abs(ymmu)/b -log(2*b); if nargout>1 dlp = sign(ymmu)/b; % dlp, derivative of log likelihood if nargout>2 % d2lp, 2nd derivative of log likelihood d2lp = zeros(size(ymmu)); if nargout>3 % d3lp, 3rd derivative of log likelihood d3lp = zeros(size(ymmu)); end end end varargout = {sum(lp),dlp,d2lp,d3lp}; else % derivative mode lp_dhyp = abs(y-mu)/b - 1; % derivative of log likelihood w.r.t. hypers d2lp_dhyp = zeros(size(mu)); % and also of the second mu derivative varargout = {lp_dhyp,d2lp_dhyp}; end case 'infEP' n = max([length(y),length(mu),length(s2),length(sn)]); on = ones(n,1); y = y.*on; mu = mu.*on; s2 = s2.*on; sn = sn.*on; % vectors only fac = 1e3; % factor between the widths of the two distributions ... % ... from when one considered a delta peak, we use 3 orders of magnitude idlik = fac*sn1 % 1st moment % M1p = sqrt(tvar).*( zp.*exp(lM0p) - exp(-zp.*zp)/sqrt(pi) ); % M1m = sqrt(tvar).*( zm.*exp(lM0m) - exp(-zm.*zm)/sqrt(pi) ); % lZmax = max([vp-lZ(id),vm-lZ(id)],[],2); % m1m0 = exp(lZmax).*( exp(vp-lZ(id)-lZmax).*M1p ... % -exp(vm-lZ(id)-lZmax).*M1m )/2 + y(id); % dlZ(id) = (m1m0-mu(id))./s2(id); % if nargout>2 % 2nd moment % M2p = 2*tvar.*exp(vp-lZ(id)+lM0p)/2+(2*tvar+sqrt(2)*tmu) ... % .*M1p.*exp(vp-lZ(id)); % M2m = 2*tvar.*exp(vm-lZ(id)+lM0m)/2+(2*tvar-sqrt(2)*tmu) ... % .*M1m.*exp(vm-lZ(id)); % m2m0 = (M2p+M2m).*sn(id)/sqrt(8) + 2*y(id).*m1m0 - y(id).^2; % d2lZ(id) = (m2m0 - m1m0.^2)./s2(id).^2 - 1./s2(id); % end % end % an implementation based on logphi(t) = log(normcdf(t)) zp = (tmu+sqrt(2)*tvar)./sqrt(tvar); zm = (tmu-sqrt(2)*tvar)./sqrt(tvar); ap = logphi(-zp)+sqrt(2)*tmu; am = logphi( zm)-sqrt(2)*tmu; lZ(id) = logsum2exp([ap,am]) + tvar - log(sn(id)*sqrt(2)); if nargout>1 lqp = -zp.^2/2 - log(2*pi)/2 - logphi(-zp); % log( N(z)/Phi(z) ) lqm = -zm.^2/2 - log(2*pi)/2 - logphi( zm); dap = -exp(lqp-log(s2(id))/2) + sqrt(2)./sn(id); dam = exp(lqm-log(s2(id))/2) - sqrt(2)./sn(id); % ( exp(ap).*dap + exp(am).*dam )./( exp(ap) + exp(am) ) dlZ(id) = expABz_expAx([ap,am],[1;1],[dap,dam],[1;1]); if nargout>2 a = sqrt(8)./sn(id)./sqrt(s2(id)); bp = 2./sn(id).^2 - (a - zp./s2(id)).*exp(lqp); bm = 2./sn(id).^2 - (a + zm./s2(id)).*exp(lqm); % d2lZ(id) = ( exp(ap).*bp + exp(am).*bm )./( exp(ap) + exp(am) ) ... % - dlZ(id).^2; d2lZ(id) = expABz_expAx([ap,am],[1;1],[bp,bm],[1;1]) - dlZ(id).^2; end end end varargout = {lZ,dlZ,d2lZ}; else % derivative mode dlZhyp = zeros(n,1); if any(idlik) dlZhyp(idlik) = 0; end if any(idgau) dlZhyp(idgau) = ... likLaplace(log(sn(idgau)), mu(idgau), y(idgau), 'infLaplace', 1); end if any(id) % substitution to obtain unit variance, zero mean Laplacian tmu = (mu(id)-y(id))./sn(id); tvar = s2(id)./sn(id).^2; zp = (tvar+tmu/sqrt(2))./sqrt(tvar); vp = tvar+sqrt(2)*tmu; zm = (tvar-tmu/sqrt(2))./sqrt(tvar); vm = tvar-sqrt(2)*tmu; dzp = (-s2(id)./sn(id)+tmu.*sn(id)/sqrt(2)) ./ sqrt(s2(id)); dvp = -2*tvar - sqrt(2)*tmu; dzm = (-s2(id)./sn(id)-tmu.*sn(id)/sqrt(2)) ./ sqrt(s2(id)); dvm = -2*tvar + sqrt(2)*tmu; lezp = lerfc(zp); % ap = exp(vp).*ezp lezm = lerfc(zm); % am = exp(vm).*ezm vmax = max([vp+lezp,vm+lezm],[],2); % subtract max to avoid numerical pb ep = exp(vp+lezp-vmax); em = exp(vm+lezm-vmax); dap = ep.*(dvp - 2/sqrt(pi)*exp(-zp.^2-lezp).*dzp); dam = em.*(dvm - 2/sqrt(pi)*exp(-zm.^2-lezm).*dzm); dlZhyp(id) = (dap+dam)./(ep+em) - 1; end varargout = {dlZhyp}; % deriv. wrt hypers end case 'infVB' if nargin<6 % variational lower site bound % t(s) = exp(-sqrt(2)|y-s|/sn) / sqrt(2*sn²) % the bound has the form: b*s - s.^2/(2*ga) - h(ga)/2 with b=y/ga!! ga = s2; n = numel(ga); b = y./ga; y = y.*ones(n,1); db = -y./ga.^2; d2b = 2*y./ga.^3; h = 2*ga/sn^2 + log(2*sn^2) + y.^2./ga; dh = 2/sn^2 - y.^2./ga.^2; d2h = 2*y.^2./ga.^3; id = ga<0; h(id) = Inf; dh(id) = 0; d2h(id) = 0; % neg. var. treatment varargout = {h,b,dh,db,d2h,d2b}; else ga = s2; dhhyp = -4*ga/sn^2 + 2; dhhyp(ga<0) = 0; % negative variances get a special treatment varargout = {dhhyp}; % deriv. wrt hyp.lik end end end % computes y = log( sum(exp(x),2) ) in a numerically safe way by subtracting % the row maximum to avoid cancelation after taking the exp % the sum is done along the rows function [y,x] = logsum2exp(logx) N = size(logx,2); max_logx = max(logx,[],2); % we have all values in the log domain, and want to calculate a sum x = exp(logx-max_logx*ones(1,N)); y = log(sum(x,2)) + max_logx; % numerically safe implementation of f(t) = log(1-erf(t)) = log(erfc(t)) function f = lerfc(t) f = zeros(size(t)); tmin = 20; tmax = 25; ok = ttmax; % evaluate tight bound interp = ~ok & ~bd; % linearly interpolate between both of them f(~ok) = log(2/sqrt(pi)) -t(~ok).^2 -log(t(~ok)+sqrt( t(~ok).^2+4/pi )); lam = 1./(1+exp( 12*(1/2-(t(interp)-tmin)/(tmax-tmin)) )); % interp. weights f(interp) = lam.*f(interp) + (1-lam).*log(erfc( t(interp) )); f(ok) = f(ok) + log(erfc( t(ok) )); % safe eval % computes y = ( (exp(A).*B)*z ) ./ ( exp(A)*x ) in a numerically safe way % The function is not general in the sense that it yields correct values for % all types of inputs. We assume that the values are close together. function y = expABz_expAx(A,x,B,z) N = size(A,2); maxA = max(A,[],2); % number of columns, max over columns A = A-maxA*ones(1,N); % subtract maximum value y = ( (exp(A).*B)*z ) ./ ( exp(A)*x ); % safe implementation of the log of phi(x) = \int_{-\infty}^x N(f|0,1) df % logphi(z) = log(normcdf(z)) function lp = logphi(z) lp = zeros(size(z)); % allocate memory zmin = -6.2; zmax = -5.5; ok = z>zmax; % safe evaluation for large values bd = z3, if norm(s2)>0, s2zero = 0; end, end % s2==0 ? if s2zero % log probability lp = -(y-mu).^2./sn2/2-log(2*pi*sn2)/2; s2 = 0; else lp = likGauss(hyp, y, mu, s2, 'infEP'); % prediction end ymu = {}; ys2 = {}; if nargout>1 ymu = mu; % first y moment if nargout>2 ys2 = s2 + sn2; % second y moment end end varargout = {lp,ymu,ys2}; else switch inf case 'infLaplace' if nargin<6 % no derivative mode if numel(y)==0, y=0; end ymmu = y-mu; dlp = {}; d2lp = {}; d3lp = {}; lp = -ymmu.^2/(2*sn2) - log(2*pi*sn2)/2; if nargout>1 dlp = ymmu/sn2; % dlp, derivative of log likelihood if nargout>2 % d2lp, 2nd derivative of log likelihood d2lp = -ones(size(ymmu))/sn2; if nargout>3 % d3lp, 3rd derivative of log likelihood d3lp = zeros(size(ymmu)); end end end varargout = {sum(lp),dlp,d2lp,d3lp}; else % derivative mode lp_dhyp = (y-mu).^2/sn2 - 1; % derivative of log likelihood w.r.t. hypers d2lp_dhyp = 2*ones(size(mu))/sn2; % and also of the second mu derivative varargout = {lp_dhyp,d2lp_dhyp}; end case 'infEP' if nargin<6 % no derivative mode lZ = -(y-mu).^2./(sn2+s2)/2 - log(2*pi*(sn2+s2))/2; % log part function dlZ = {}; d2lZ = {}; if nargout>1 dlZ = (y-mu)./(sn2+s2); % 1st derivative w.r.t. mean if nargout>2 d2lZ = -1./(sn2+s2); % 2nd derivative w.r.t. mean end end varargout = {lZ,dlZ,d2lZ}; else % derivative mode dlZhyp = ((y-mu).^2./(sn2+s2)-1) ./ (1+s2./sn2); % deriv. w.r.t. hyp.lik varargout = {dlZhyp}; end case 'infVB' if nargin<6 % variational lower site bound % t(s) = exp(-(y-s)^2/2sn2)/sqrt(2*pi*sn2) % the bound has the form: b*s - s.^2/(2*ga) - h(ga)/2 with b=y/ga ga = s2; n = numel(ga); b = y./ga; y = y.*ones(n,1); db = -y./ga.^2; d2b = 2*y./ga.^3; h = zeros(n,1); dh = h; d2h = h; % allocate memory for return args id = ga(:)<=sn2+1e-8; % OK below noise variance h(id) = y(id).^2./ga(id) + log(2*pi*sn2); h(~id) = Inf; dh(id) = -y(id).^2./ga(id).^2; d2h(id) = 2*y(id).^2./ga(id).^3; id = ga<0; h(id) = Inf; dh(id) = 0; d2h(id) = 0; % neg. var. treatment varargout = {h,b,dh,db,d2h,d2b}; else ga = s2; n = numel(ga); dhhyp = zeros(n,1); dhhyp(ga(:)<=sn2) = 2; dhhyp(ga<0) = 0; % negative variances get a special treatment varargout = {dhhyp}; % deriv. w.r.t. hyp.lik end end end gpml-matlab-v3.1-2010-09-27/lik/likErf.m000644 000765 000024 00000013105 11422240322 017221 0ustar00carlstaff000000 000000 function [varargout] = likErf(hyp, y, mu, s2, inf, i) % likErf - Error function or cumulative Gaussian likelihood function for binary % classification or probit regression. The expression for the likelihood is % likErf(t) = (1+erf(t/sqrt(2)))/2 = normcdf(t). % % Several modes are provided, for computing likelihoods, derivatives and moments % respectively, see likelihoods.m for the details. In general, care is taken % to avoid numerical issues when the arguments are extreme. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-07-22. % % See also likFunctions.m. if nargin<2, varargout = {'0'}; return; end % report number of hyperparameters if nargin>1, y = sign(y); y(y==0) = 1; else y = 1; end % allow only +/- 1 values if numel(y)==0, y = 1; end if nargin<5 % prediction mode if inf is not present y = y.*ones(size(mu)); % make y a vector s2zero = 1; if nargin>3, if norm(s2)>0, s2zero = 0; end, end % s2==0 ? if s2zero % log probability evaluation [p,lp] = cumGauss(y,mu); else % prediction lp = likErf(hyp, y, mu, s2, 'infEP'); p = exp(lp); end ymu = {}; ys2 = {}; if nargout>1 ymu = 2*p-1; % first y moment if nargout>2 ys2 = 4*p.*(1-p); % second y moment end end varargout = {lp,ymu,ys2}; else % inference mode switch inf case 'infLaplace' if nargin<6 % no derivative mode f = mu; yf = y.*f; % product latents and labels dlp = {}; d2lp = {}; d3lp = {}; % return arguments [p,lp] = cumGauss(y,f); if nargout>1 % derivative of log likelihood n_p = gauOverCumGauss(yf,p); dlp = y.*n_p; % derivative of log likelihood if nargout>2 % 2nd derivative of log likelihood d2lp = -n_p.^2 - yf.*n_p; if nargout>3 % 3rd derivative of log likelihood d3lp = 2*y.*n_p.^3 +3*f.*n_p.^2 +y.*(f.^2-1).*n_p; end end end varargout = {sum(lp),dlp,d2lp,d3lp}; else % derivative mode varargout = {[]}; % derivative w.r.t. hypers end case 'infEP' if nargin<6 % no derivative mode z = mu./sqrt(1+s2); dlZ = {}; d2lZ = {}; [junk,lZ] = cumGauss(y,z); % log part function if numel(y)>0, z=z.*y; end if nargout>1 if numel(y)==0, y=1; end n_p = gauOverCumGauss(z,exp(lZ)); dlZ = y.*n_p./sqrt(1+s2); % 1st derivative wrt mean if nargout>2 d2lZ = -n_p.*(z+n_p)./(1+s2); % 2nd derivative wrt mean end end varargout = {lZ,dlZ,d2lZ}; else % derivative mode varargout = {[]}; % deriv. wrt hyp.lik end case 'infVB' % naive variational lower site bound based on asymptotical properties of lik % normcdf(t) -> -(t²-2dt+c)/2 for t->-oo (tight lower bound) d = 0.158482605320942; c = -1.785873318175113; ga = s2; n = numel(ga); b = d*y.*ones(n,1); db = zeros(n,1); d2b = db; h = -2*c*ones(n,1); h(ga>1) = Inf; dh = zeros(n,1); d2h = dh; varargout = {h,b,dh,db,d2h,d2b}; end end function [p,lp] = cumGauss(y,f) if numel(y)>0, yf = y.*f; else yf = f; end % product of latents and labels p = (1+erf(yf/sqrt(2)))/2; % likelihood if nargout>1, lp = logphi(yf,p); end % log likelihood % safe implementation of the log of phi(x) = \int_{-\infty}^x N(f|0,1) df % logphi(z) = log(normcdf(z)) function lp = logphi(z,p) lp = zeros(size(z)); % allocate memory zmin = -6.2; zmax = -5.5; ok = z>zmax; % safe evaluation for large values bd = z-5; % naive evaluation for large values of f n_p(ok) = (exp(-f(ok).^2/2)/sqrt(2*pi)) ./ p(ok); bd = f<-6; % tight upper bound evaluation n_p(bd) = sqrt(f(bd).^2/4+1)-f(bd)/2; interp = ~ok & ~bd; % linearly interpolate between both of them tmp = f(interp); lam = -5-f(interp); n_p(interp) = (1-lam).*(exp(-tmp.^2/2)/sqrt(2*pi))./p(interp) + ... lam .*(sqrt(tmp.^2/4+1)-tmp/2); gpml-matlab-v3.1-2010-09-27/lik/likLogistic.m000644 000765 000024 00000017346 11422240322 020275 0ustar00carlstaff000000 000000 function [varargout] = likLogistic(hyp, y, mu, s2, inf, i) % likLogistic - logistic function for binary classification or logit regression. % The expression for the likelihood is % likLogistic(t) = 1./(1+exp(-t)). % % Several modes are provided, for computing likelihoods, derivatives and moments % respectively, see likelihoods.m for the details. In general, care is taken % to avoid numerical issues when the arguments are extreme. The moments % \int f^k likLogistic(y,f) N(f|mu,var) df are calculated via a cumulative % Gaussian scale mixture approximation. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-07-22. % % See also likFunctions.m. if nargin<2, varargout = {'0'}; return; end % report number of hyperparameters if nargin>1, y = sign(y); y(y==0) = 1; else y = 1; end % allow only +/- 1 values if numel(y)==0, y = 1; end if nargin<5 % prediction mode if inf is not present y = y.*ones(size(mu)); % make y a vector s2zero = 1; if nargin>3, if norm(s2)>0, s2zero = 0; end, end % s2==0 ? if s2zero % log probability evaluation yf = y.*mu; % product latents and labels lp = yf; ok = -351 p = exp(lp); ymu = 2*p-1; % first y moment if nargout>2 ys2 = 4*p.*(1-p); % second y moment end end varargout = {lp,ymu,ys2}; else % inference mode switch inf case 'infLaplace' if nargin<6 % no derivative mode f = mu; yf = y.*f; s = -yf; % product latents and labels dlp = {}; d2lp = {}; d3lp = {}; % return arguments ps = max(0,s); lp = -(ps+log(exp(-ps)+exp(s-ps))); % lp = -(log(1+exp(s))) if nargout>1 % first derivatives s = min(0,f); p = exp(s)./(exp(s)+exp(s-f)); % p = 1./(1+exp(-f)) dlp = (y+1)/2-p; % derivative of log likelihood if nargout>2 % 2nd derivative of log likelihood d2lp = -exp(2*s-f)./(exp(s)+exp(s-f)).^2; if nargout>3 % 3rd derivative of log likelihood d3lp = 2*d2lp.*(0.5-p); end end end varargout = {sum(lp),dlp,d2lp,d3lp}; else % derivative mode varargout = {[]}; % derivative w.r.t. hypers end case 'infEP' if nargin<6 % no derivative mode y = y.*ones(size(mu)); % make y a vector % likLogistic(t) \approx 1/2 + \sum_{i=1}^5 (c_i/2) erf(lam_i/sqrt(2)t) lam = sqrt(2)*[0.44 0.41 0.40 0.39 0.36]; % approx coeffs lam_i and c_i c = [1.146480988574439e+02; -1.508871030070582e+03; 2.676085036831241e+03; -1.356294962039222e+03; 7.543285642111850e+01 ]; [lZc,dlZc,d2lZc] = likErf([], y*ones(1,5), mu*lam, s2*(lam.^2), inf); lZ = log_expA_x(lZc,c); % A=lZc, B=dlZc, d=c.*lam', lZ=log(exp(A)*c) dlZ = expABz_expAx(lZc, c, dlZc, c.*lam'); % ((exp(A).*B)*d)./(exp(A)*c) % d2lZ = ((exp(A).*Z)*e)./(exp(A)*c) - dlZ.^2 where e = c.*(lam.^2)' d2lZ = expABz_expAx(lZc, c, dlZc.^2+d2lZc, c.*(lam.^2)') - dlZ.^2; % The scale mixture approximation does not capture the correct asymptotic % behavior; we have linear decay instead of quadratic decay as suggested % by the scale mixture approximation. By observing that for large values % of -f*y ln(p(y|f)) of likLogistic is linear in f with slope y, we are % able to analytically integrate the tail region; there is no contribution % to the second derivative val = abs(mu)-196/200*s2-4; % empirically determined bound at val==0 lam = 1./(1+exp(-10*val)); % interpolation weights lZtail = min(s2/2-abs(mu),-.1); % apply the same to p(y|f) = 1 - p(-y|f) dlZtail = -sign(mu); id = y.*mu>0; lZtail(id) = log(1-exp(lZtail(id))); % label and mean agree dlZtail(id) = 0; lZ = (1-lam).* lZ + lam.* lZtail; % interpolate between scale mixture .. dlZ = (1-lam).*dlZ + lam.*dlZtail; % .. and tail approximation varargout = {lZ,dlZ,d2lZ}; else % derivative mode varargout = {[]}; % deriv. wrt hyp.lik end case 'infVB' % variational lower site bound % using -log(1+exp(-s)) = s/2 -log( 2*cosh(s/2) ); % the bound has the form: b*s - s.^2/(2*ga) - h(ga)/2 with b=1/2 ga = s2; n = numel(ga); b = (y/2).*ones(n,1); db = zeros(n,1); d2b = db; h = zeros(n,1); dh = h; d2h = h; % allocate memory for return argument idlo = ga(:)<=4; h(idlo) = 2*log(2); % constant value below 4 idup = 50<=ga(:); zn = zeros(n,1); % linear behavior for large gammas h(idup) = ga(idup)/4; dh(idup) = zn(idup)+1/4; d2h(idup) = zn(idup); id = ~idlo & ~idup; % interesting zone is in between [g,dg,d2g] = inv_xcothx(1/4*ga(id)); thg = tanh(g); h(id) = 2*logcosh(g) +2*log(2) -g.*thg; g1mt2 = g.*(1-thg.^2); dh(id) = dg.*( thg - g1mt2 )/4; d2h(id) = ( g1mt2.*(2*dg.^2.*thg-d2g) + d2g.*thg )/16; varargout = {h,b,dh,db,d2h,d2b}; end end % numerically safe version of log(cosh(x)) = log(exp(x)+exp(-x))-log(2) function f = logcosh(x) f = abs(x) + log(1+exp(-2*abs(x))) - log(2); % Invert f(x) = x.*coth(x) = y, return the positive value % uses Newton's method to minimize (y-f(x))² w.r.t. x function [x,dx,d2x] = inv_xcothx(y) % f = @(x) x.*coth(x); % df = @(x) x + (1-f(x)).*coth(x); % d2f = @(x) 2*(1-f(x)).*(1-coth(x).^2); if numel(y)==0, x=[]; dx=[]; d2x=[]; return, end x = sqrt(y.^2-1); i = 1; % init ep = eps./(1+abs(x)); th=tanh(x); fx=(x+ep)./(th+ep); % function r = fx-y; % init while i==1 || (i<10 && max(abs(r))>1e-12) dfx = x + (1-fx)./(th+ep); % first derivative d2fx = 2*( 1-fx + (x-th+ep/3)./(th.*th.*th+ep) ); % second derivative x = x - r.*dfx./(dfx.^2+r.*d2fx+ep); % Newton step ep = eps./(1+abs(x)); th=tanh(x); fx=(x+ep)./(th+ep); r = fx-y; i = i+1; end % first derivative dx/dy % derivatives of inverse functions are reciprocals dx = 1./dfx; % second derivative d2x/dy2 % quotient rule and chaine rule d2x = -d2fx.*dx./dfx./dfx; % computes y = log( exp(A)*x ) in a numerically safe way by subtracting the % maximal value in each row to avoid cancelation after taking the exp function y = log_expA_x(A,x) N = size(A,2); maxA = max(A,[],2); % number of columns, max over columns y = log(exp(A-maxA*ones(1,N))*x) + maxA; % exp(A) = exp(A-max(A))*exp(max(A)) % computes y = ( (exp(A).*B)*z ) ./ ( exp(A)*x ) in a numerically safe way % The function is not general in the sense that it yields correct values for % all types of inputs. We assume that the values are close together. function y = expABz_expAx(A,x,B,z) N = size(A,2); maxA = max(A,[],2); % number of columns, max over columns A = A-maxA*ones(1,N); % subtract maximum value y = ( (exp(A).*B)*z ) ./ ( exp(A)*x ); gpml-matlab-v3.1-2010-09-27/inf/infEP.m000644 000765 000024 00000013652 11474502347 017031 0ustar00carlstaff000000 000000 function [post nlZ dnlZ] = infEP(hyp, mean, cov, lik, x, y) % Expectation Propagation approximation to the posterior Gaussian Process. % The function takes a specified covariance function (see covFunction.m) and % likelihood function (see likFunction.m), and is designed to be used with % gp.m. See also infFunctions.m. In the EP algorithm, the sites are % updated in random order, for better performance when cases are ordered % according to the targets. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-02-25. % % See also INFMETHODS.M. persistent last_ttau last_tnu % keep tilde parameters between calls tol = 1e-4; max_sweep = 10; min_sweep = 2; % tolerance to stop EP iterations if ~ischar(lik), lik = func2str(lik); end % make likelihood variable a string inf = 'infEP'; n = size(x,1); K = feval(cov{:}, hyp.cov, x); % evaluate the covariance matrix m = feval(mean{:}, hyp.mean, x); % evaluate the mean vector % A note on naming: variables are given short but descriptive names in % accordance with Rasmussen & Williams "GPs for Machine Learning" (2006): mu % and s2 are mean and variance, nu and tau are natural parameters. A leading t % means tilde, a subscript _ni means "not i" (for cavity parameters), or _n % for a vector of cavity parameters. % marginal likelihood for ttau = tnu = zeros(n,1); equals n*log(2) for likCum* nlZ0 = -sum(feval(lik, hyp.lik, y, m, diag(K), inf)); if any(size(last_ttau) ~= [n 1]) % find starting point for tilde parameters ttau = zeros(n,1); % initialize to zero if we have no better guess tnu = zeros(n,1); Sigma = K; % initialize Sigma and mu, the parameters of .. mu = zeros(n,1); % .. the Gaussian posterior approximation nlZ = nlZ0; else ttau = last_ttau; % try the tilde values from previous call tnu = last_tnu; [Sigma, mu, nlZ, L] = epComputeParams(K, y, ttau, tnu, lik, hyp, m, inf); if nlZ > nlZ0 % if zero is better .. ttau = zeros(n,1); % .. then initialize with zero instead tnu = zeros(n,1); Sigma = K; % initialize Sigma and mu, the parameters of .. mu = zeros(n,1); % .. the Gaussian posterior approximation nlZ = nlZ0; end end nlZ_old = Inf; sweep = 0; % converged, max. sweeps or min. sweeps? while (abs(nlZ-nlZ_old) > tol && sweep < max_sweep) || sweep2 % do we want derivatives? dnlZ = hyp; % allocate space for derivatives ssi = sqrt(ttau); V = L'\(repmat(ssi,1,n).*K); Sigma = K - V'*V; mu = Sigma*tnu; tau_n = 1./diag(Sigma)-ttau; % compute the log marginal likelihood nu_n = mu./diag(Sigma)-tnu; % vectors of cavity parameters F = alpha*alpha'-repmat(sW,1,n).*solve_chol(L,diag(sW)); % covariance hypers for j=1:length(hyp.cov) dK = feval(cov{:}, hyp.cov, x, [], j); dnlZ.cov(j) = -sum(sum(F.*dK))/2; end for i = 1:numel(hyp.lik) % likelihood hypers dlik = feval(lik, hyp.lik, y, nu_n./tau_n+m, 1./tau_n, inf, i); dnlZ.lik(i) = -sum(dlik); end [junk,dlZ] = feval(lik, hyp.lik, y, nu_n./tau_n+m, 1./tau_n, inf); % mean hyps for i = 1:numel(hyp.mean) dm = feval(mean{:}, hyp.mean, x, i); dnlZ.mean(i) = -dlZ'*dm; end end % function to compute the parameters of the Gaussian approximation, Sigma and % mu, and the negative log marginal likelihood, nlZ, from the current site % parameters, ttau and tnu. Also returns L (useful for predictions). function [Sigma mu nlZ L] = epComputeParams(K, y, ttau, tnu, lik, hyp, m, inf) n = length(y); % number of training cases ssi = sqrt(ttau); % compute Sigma and mu L = chol(eye(n)+ssi*ssi'.*K); % L'*L=B=eye(n)+sW*K*sW V = L'\(repmat(ssi,1,n).*K); Sigma = K - V'*V; mu = Sigma*tnu; tau_n = 1./diag(Sigma)-ttau; % compute the log marginal likelihood nu_n = mu./diag(Sigma)-tnu+m.*tau_n; % vectors of cavity parameters lZ = feval(lik, hyp.lik, y, nu_n./tau_n, 1./tau_n, inf); nlZ = sum(log(diag(L))) -sum(lZ) -tnu'*Sigma*tnu/2 ... -(nu_n-m.*tau_n)'*((ttau./tau_n.*(nu_n-m.*tau_n)-2*tnu)./(ttau+tau_n))/2 ... +sum(tnu.^2./(tau_n+ttau))/2-sum(log(1+ttau./tau_n))/2; gpml-matlab-v3.1-2010-09-27/inf/infFITC.m000644 000765 000024 00000007712 11474502347 017252 0ustar00carlstaff000000 000000 function [post nlZ dnlZ] = infFITC(hyp, mean, cov, lik, x, y) % FITC approximation to the posterior Gaussian process. The function is % equivalent to infExact with the covariance function: % Kt = Q + G; G = diag(diag(K-Q); Q = Ku'*inv(Kuu + snu2*eye(nu))*Ku; % where Ku and Kuu are covariances w.r.t. to inducing inputs xu and % snu2 = sn2/1e6 is the noise of the inducing inputs. We fixed the standard % deviation of the inducing inputs to be a one per mil of the measurement noise % standard deviation. % The implementation exploits the Woodbury matrix identity % inv(Kt) = inv(G) - inv(G)*Ku'*inv(Kuu+Ku*inv(G)*Ku')*Ku*inv(G) % in order to be applicable to large datasets. The computational complexity % is O(n nu^2) where n is the number of data points x and nu the number of % inducing inputs in xu. % The function takes a specified covariance function (see covFunction.m) and % likelihood function (see likFunction.m), and is designed to be used with % gp.m and in conjunction with covFITC and likGauss. % % Copyright (c) by Ed Snelson, Carl Edward Rasmussen % and Hannes Nickisch, 2010-11-26. % % See also INFMETHODS.M, COVFITC.M. likstr = lik; if ~ischar(lik), likstr = func2str(lik); end if ~strcmp(likstr,'likGauss') % NOTE: no explicit call to likGauss error('FITC inference only possible with Gaussian likelihood'); end cov1 = cov{1}; if isa(cov1, 'function_handle'), cov1 = func2str(cov1); end if ~strcmp(cov1,'covFITC'); error('Only covFITC supported.'), end % check cov [diagK,Kuu,Ku] = feval(cov{:}, hyp.cov, x); % evaluate covariance matrix m = feval(mean{:}, hyp.mean, x); % evaluate mean vector [n, D] = size(x); nu = size(Kuu,1); sn2 = exp(2*hyp.lik); % noise variance of likGauss snu2 = 1e-6*sn2; % hard coded inducing inputs noise Luu = chol(Kuu + snu2*eye(nu)); % Kuu + snu2*I = Luu'*Luu V = Luu'\Ku; % V = inv(Luu')*Ku => V'*V = Q dg = diagK + sn2 - sum(V.*V,1)'; % D + sn2*eye(n) = diag(K) + sn2 - diag(Q) V = V./repmat(sqrt(dg)',nu,1); Lu = chol(eye(nu) + V*V'); r = (y-m)./sqrt(dg); be = Lu'\(V*r); iKuu = solve_chol(Luu,eye(nu)); % inv(Kuu + snu2*I) = iKuu post.alpha = Luu\(Lu\be); % return the posterior parameters post.L = solve_chol(Lu*Luu,eye(nu)) - iKuu; % Sigma-inv(Kuu) post.sW = []; % unused for FITC if nargout>1 % do we want the marginal likelihood nlZ = sum(log(diag(Lu))) + (sum(log(dg)) + n*log(2*pi) + r'*r - be'*be)/2; if nargout>2 % do we want derivatives? dnlZ = hyp; % allocate space for derivatives W = Ku./repmat(sqrt(dg)',nu,1); W = chol(Kuu+W*W'+snu2*eye(nu))'\Ku; % inv(K) = inv(G) - inv(G)*W'*W*inv(G); al = (y-m - W'*(W*((y-m)./dg)))./dg; B = iKuu*Ku; Wdg = W./repmat(dg',nu,1); w = B*al; for i = 1:numel(hyp.cov) [ddiagKi,dKuui,dKui] = feval(cov{:}, hyp.cov, x, [], i); % eval cov deriv R = 2*dKui-dKuui*B; v = ddiagKi - sum(R.*B,1)'; % diag part of cov deriv dnlZ.cov(i) = (ddiagKi'*(1./dg) +w'*(dKuui*w-2*(dKui*al)) -al'*(v.*al) ... - sum(Wdg.*Wdg,1)*v - sum(sum((R*Wdg').*(B*Wdg'))) )/2; end dnlZ.lik = sn2*(sum(1./dg) -sum(sum(W.*W,1)'./(dg.*dg)) -al'*al); % since snu2 is a fixed fraction of sn2, there is a covariance-like term in % the derivative as well dKuui = 2*snu2; R = -dKuui*B; v = -sum(R.*B,1)'; % diag part of cov deriv dnlZ.lik = dnlZ.lik + (w'*dKuui*w -al'*(v.*al)... - sum(Wdg.*Wdg,1)*v - sum(sum((R*Wdg').*(B*Wdg'))) )/2; for i = 1:numel(hyp.mean) dnlZ.mean(i) = -feval(mean{:}, hyp.mean, x, i)'*al; end end endgpml-matlab-v3.1-2010-09-27/inf/infVB.m000644 000765 000024 00000013522 11447142675 017034 0ustar00carlstaff000000 000000 function [post, nlZ, dnlZ] = infVB(hyp, mean, cov, lik, x, y) % Variational approximation to the posterior Gaussian process with MKL % covariance function hyperparameter optimisation. % The function takes a likelihood function (see likFunction.m), and is designed % to be used with gp.m. See also infFunctions.m. % % Minimisation of an upper bound on the negative marginal likelihood % \int N(f|0,K) p(y|f) df \ge Psi(ga,theta) with hyperparameters % theta=cov.hyp and likelihood lik(y,f) = p(y|f). % % Psi(ga,theta) = nlZ = % = (ln|K+ga|-ln|ga| + h(ga) - b'*inv(inv(K)+inv(ga)))*b)/2 % = (ln|A| + ln|K| + h(ga) - b'*inv(A)*b)/2 with A = inv(K)+inv(ga) % % The map (ga,K) |-> Psi(ga,theta) - ln|K| is jointly convex and ln|K| is % concave in its linear parameters theta. % % We optimise the convex Psi(ga,theta)-ln|K|+z'*theta/2 + B(t,theta) criterion % in the outer loop using the barrier function B(t,theta) = -sum(log(theta))/t % to enforce positivity w.r.t. theta. % The linear term z'*theta/2 is an upper bound on the concave ln|K| term. % We use an interleaved optimisation with inner loops doing Newton steps in ga % and an outer loop doing joint Newton updates in (ga,theta). Line searches are % done using derivative-free 'Brent's minimum' search. % % Copyright (c) by Hannes Nickisch 2010-07-22. % % See also INFMETHODS.M. maxitinner = 20; % number of inner Newton steps in ga % some less important parameters ep = 1e-8; % small constant for the ridge added to the stab. Newton direction smax = 5; Nline = 15; thr = 1e-4; % line search parameters tol = 1e-7; % tolerance for when to stop the Newton iterations inf = 'infVB'; [n,D] = size(x); K = feval(cov{:}, hyp.cov, x); % evaluate the covariance matrix m = feval(mean{:}, hyp.mean, x); % evaluate the mean vector if ~ischar(lik), lik = func2str(lik); end if (norm(m)>1e-10 || numel(hyp.mean)>0) ... && (strcmp(lik,'likErf')||strcmp(lik,'likLogistic')) error('only meanZero implemented for classification') end y = y-m; % no we have either zero mean or non-classification likelihoods if strcmp(lik,'likGauss') ga = exp(2*hyp.lik)*ones(n,1); % best ga is known for Gaussian likelihood elseif strcmp(lik,'likErf') ga = ones(n,1); % use a fixed ga for errf likelihood due to asymptotics else % INNER compute the Newton direction of ga ga = ones(n,1); % initial values itinner = 0; nlZ_new = 1e100; nlZ_old = Inf; % make sure while loop starts while nlZ_old-nlZ_new>tol && itinner0 Psi_s = @(s,ddga,ga,K,inf,hyp,lik,y) Psi(ga+s*ddga,K,inf,hyp,lik,y); [s,nlZ_new] = brentmin(0,s,Nline,thr, Psi_s,0,ddga,ga,K,inf,hyp,lik,y); ga = abs(ga + s*ddga); % update variational parameters end end [nlZ,dnlZ,d2nlZ,b] = Psi(ga,K,inf,hyp,lik,y); % upp bd on neg log marg lik W = 1./ga; sW = sqrt(W); % return the posterior parameters L = chol(eye(n)+sW*sW'.*K); % L'*L = B =eye(n)+sW*K*sW iKtil = repmat(sW,1,n).*solve_chol(L,diag(sW)); % sW*B^-1*sW=inv(K+inv(W)) alpha = b - iKtil*(K*b); post.alpha = alpha; post.sW = sW; post.L = L; if nargout>2 % do we want derivatives? dnlZ = hyp; % allocate space for derivatives for j=1:length(hyp.cov) % covariance hypers dK = feval(cov{:}, hyp.cov, x, [], j); if j==1, v = iKtil*(b./W); end dnlZ.cov(j) = sum(sum(iKtil.*dK))/2 - (v'*dK*v)/2; % implicit derivative = 0 end if ~strcmp(lik,'likGauss') % likelihood hypers for j=1:length(hyp.lik) dhhyp = feval(lik,hyp.lik,y,[],ga,inf,j); dnlZ.lik(j) = sum(dhhyp)/2; % implicit derivative = 0 end else % special treatment for the Gaussian case dnlZ.lik = sum(sum( (L'\eye(n)).^2 )) - exp(2*hyp.lik)*(alpha'*alpha); end for j=1:length(hyp.mean) % mean hypers dm = feval(mean{:}, hyp.mean, x, j); dnlZ.mean(j) = -alpha'*dm; % implicit derivative = 0 end end % variational lower bound along with derivatives w.r.t. ga and theta=hyp.cov % psi = (ln|K+ga|-ln|ga| + h(ga) - b'*inv(inv(K)+inv(ga)))*b)/2 % = (ln|A| + ln|K| + h(ga) - b'*inv(A)*b)/2 with A = inv(K)+inv(ga) % the code is numerically stable function [nlZ,dga,d2ga,b] = Psi(ga,K,inf,hyp,lik,y) n = size(K,1); [h,b,dh,db,d2h,d2b] = feval(lik,hyp.lik,y,[],ga,inf); W = 1./ga; sW = sqrt(W); L = chol(eye(n)+sW*sW'.*K); % sum(log(diag(L))) = log|B|/2 C = L'\(repmat(sW,1,n).*K); t = C*b; % t'*t-b'*K*b = -b'*inv(inv(K)+diag(1./ga))*b nlZ = sum(log(diag(L))) + ( sum(h) + t'*t-b'*K*b )/2; if nargout>1 % Hessian w.r.t. variational parameters gamma iKtil = repmat(sW,1,n).*solve_chol(L,diag(sW)); % sW*inv(B)*sW=inv(K+inv(W)) Khat = K-C'*C; v = Khat*b; % K-K*sW*inv(B)*sW*K=inv(inv(K)+W) dga = ( diag(iKtil)-1./ga + dh - (v./ga).^2 - 2*v.*db )/2; if nargout>2 % gradient w.r.t. variational parameters gamma w = v./ga.^2; d2ga = ( -iKtil.^2+diag(1./ga.^2) + diag(d2h) )/2 ... -Khat.*(w*(w+2*db)') -Khat.*(db*db') +diag(w.^2.*ga)-diag(v.*d2b); d2ga = (d2ga+d2ga')/2; % symmetrise end end gpml-matlab-v3.1-2010-09-27/inf/infLaplace.m000644 000765 000024 00000015253 11504370467 020065 0ustar00carlstaff000000 000000 function [post nlZ dnlZ] = infLaplace(hyp, mean, cov, lik, x, y) % Laplace approximation to the posterior Gaussian process. % The function takes a specified covariance function (see covFunction.m) and % likelihood function (see likFunction.m), and is designed to be used with % gp.m. See also infFunctions.m. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-02-23. % % See also INFMETHODS.M. persistent last_alpha % copy of the last alpha tol = 1e-6; % tolerance for when to stop the Newton iterations smax = 2; Nline = 10; thr = 1e-4; % line search parameters maxit = 20; % max number of Newton steps in f if ~ischar(lik), lik = func2str(lik); end % make likelihood variable a string inf = 'infLaplace'; n = size(x,1); K = feval(cov{:}, hyp.cov, x); % evaluate the covariance matrix m = feval(mean{:}, hyp.mean, x); % evaluate the mean vector Psi_old = Inf; % make sure while loop starts by the largest old objective val if any(size(last_alpha)~=[n,1]) % find a good starting point for alpha and f alpha = zeros(n,1); f = K*alpha+m; % start at mean if sizes do not match [lp,dlp,d2lp] = feval(lik,hyp.lik,y,f,[],inf); W = -d2lp; Psi_new = -lp; else alpha = last_alpha; f = K*alpha+m; % try last one [lp,dlp,d2lp] = feval(lik,hyp.lik,y,f,[],inf); W = -d2lp; Psi_new = alpha'*(f-m)/2 - lp; % objective for last alpha Psi_def = -feval(lik,hyp.lik,y,m,[],inf); % objective for default init f==m if Psi_def < Psi_new % if default is better, we use it alpha = zeros(n,1); f = K*alpha+m; [lp,dlp,d2lp] = feval(lik,hyp.lik,y,f,[],inf); W = -d2lp; Psi_new = -lp; end end isWneg = any(W<0); % flag indicating whether we found negative values of W it = 0; % this happens for the Student's t likelihood while Psi_old - Psi_new > tol && it2 % do we want derivatives? dnlZ = hyp; % allocate space for derivatives if isWneg % switch between Cholesky and LU decomposition mode Z = -post.L; % inv(K+inv(W)) g = sum(iA.*K,2)/2; % deriv. of ln|B| wrt W; g = diag(inv(inv(K)+diag(W)))/2 else Z = repmat(sW,1,n).*solve_chol(post.L,diag(sW)); %sW*inv(B)*sW=inv(K+inv(W)) C = post.L'\(repmat(sW,1,n).*K); % deriv. of ln|B| wrt W g = (diag(K)-sum(C.^2,1)')/2; % g = diag(inv(inv(K)+W))/2 end dfhat = g.*d3lp; % deriv. of nlZ wrt. fhat: dfhat=diag(inv(inv(K)+W)).*d3lp/2 for j=1:length(hyp.cov) % covariance hypers dK = feval(cov{:}, hyp.cov, x, [], j); dnlZ.cov(j) = sum(sum(Z.*dK))/2 - alpha'*dK*alpha/2; % explicit part b = dK*dlp; % b-K*(Z*b) = inv(eye(n)+K*diag(W))*b dnlZ.cov(j) = dnlZ.cov(j) - dfhat'*( b-K*(Z*b) ); % implicit part end for j=1:length(hyp.lik) % likelihood hypers [lp_dhyp,d2lp_dhyp] = feval(lik,hyp.lik,y,f,[],inf,j); dnlZ.lik(j) = -g'*d2lp_dhyp - sum(lp_dhyp); % explicit part, implicit is 0 end for j=1:length(hyp.mean) % mean hypers dm = feval(mean{:}, hyp.mean, x, j); dnlZ.mean(j) = -alpha'*dm; % explicit part dnlZ.mean(j) = dnlZ.mean(j) - dfhat'*(dm-K*(Z*dm)); % implicit part end end % criterion Psi at alpha + s*dalpha for line search function [Psi,alpha,f,dlp,W] = Psi_line(s,dalpha,alpha,hyp,K,m,lik,y,inf) alpha = alpha + s*dalpha; f = K*alpha+m; [lp,dlp,d2lp] = feval(lik,hyp.lik,y,f,[],inf); W = -d2lp; Psi = alpha'*(f-m)/2 - lp; % Compute the log determinant ldA and the inverse iA of a square nxn matrix % A = eye(n) + K*diag(w) from its LU decomposition; for negative definite A, we % return ldA = Inf. We also return mwiA = -diag(w)*inv(A). function [ldA,iA,mwiA] = logdetA(K,w) [m,n] = size(K); if m~=n, error('K has to be nxn'), end A = eye(n)+K.*repmat(w',n,1); [L,U,P] = lu(A); u = diag(U); % compute LU decomposition, A = P'*L*U signU = prod(sign(u)); % sign of U detP = 1; % compute sign (and det) of the permutation matrix P p = P*(1:n)'; for i=1:n if i~=p(i) detP = -detP; j = find(p==i); p([i,j]) = p([j,i]); % swap entries end end if signU~=detP % log becomes complex for negative values, encoded by infinity ldA = Inf; else % det(L) = 1 and U triangular => det(A) = det(P)*prod(diag(U)) ldA = sum(log(abs(u))); end if nargout>1, iA = inv(U)*inv(L)*P; end % return the inverse, as well if nargout>2, mwiA = -repmat(w,1,n).*iA; end gpml-matlab-v3.1-2010-09-27/inf/infExact.m000644 000765 000024 00000003511 11527514620 017556 0ustar00carlstaff000000 000000 function [post nlZ dnlZ] = infExact(hyp, mean, cov, lik, x, y) % Exact inference for a GP with Gaussian likelihood. Compute a parametrization % of the posterior, the negative log marginal likelihood and its derivatives % w.r.t. the hyperparameters. See also "help infMethods". % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2011-02-18 % % See also INFMETHODS.M. likstr = lik; if ~ischar(lik), likstr = func2str(lik); end if ~strcmp(likstr,'likGauss') % NOTE: no explicit call to likGauss error('Exact inference only possible with Gaussian likelihood'); end [n, D] = size(x); K = feval(cov{:}, hyp.cov, x); % evaluate covariance matrix m = feval(mean{:}, hyp.mean, x); % evaluate mean vector sn2 = exp(2*hyp.lik); % noise variance of likGauss L = chol(K/sn2+eye(n)); % Cholesky factor of covariance with noise alpha = solve_chol(L,y-m)/sn2; post.alpha = alpha; % return the posterior parameters post.sW = ones(n,1)/sqrt(sn2); % sqrt of noise precision vector post.L = L; % L = chol(eye(n)+sW*sW'.*K) if nargout>1 % do we want the marginal likelihood? nlZ = (y-m)'*alpha/2 + sum(log(diag(L))) + n*log(2*pi*sn2)/2; % -log marg lik if nargout>2 % do we want derivatives? dnlZ = hyp; % allocate space for derivatives Q = solve_chol(L,eye(n))/sn2 - alpha*alpha'; % precompute for convenience for i = 1:numel(hyp.cov) dnlZ.cov(i) = sum(sum(Q.*feval(cov{:}, hyp.cov, x, [], i)))/2; end dnlZ.lik = sn2*trace(Q); for i = 1:numel(hyp.mean), dnlZ.mean(i) = -feval(mean{:}, hyp.mean, x, i)'*alpha; end end end gpml-matlab-v3.1-2010-09-27/inf/infLOO.m000644 000765 000024 00000010217 11504370467 017150 0ustar00carlstaff000000 000000 function [post nloo dnloo loo] = infLOO(hyp, mean, cov, lik, x, y) % Leave-One-Out - Perform Least-Squares GP predictions in the style of the % Pseudo Likelihood in §5.4.2 p. 117 and the Probabilistic Least-squares % Classifier of §6.5 p. 146 of the GPML book. % For Gaussian likelihood, the prediction is the same as with infExact, for % lik{Erf,Logistic} the result corresponds to probabilistic least-squares % classification. In any case, the objective returned is not the marginal % likelihood but the predictive log probability. % The standard deviation of the noise either comes from the sn paramtere of the % likelihood function lik{Gauss,Laplace,Sech2} or from the covariance function % by means of using covSum and covNoise i.e. for lik{Erf,Logistic}. % We then compute the negative leave-one-out predictive probability and its % derivatives w.r.t. the hyperparameters. See also "help infMethods". % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-12-14 % % See also INFMETHODS.M. [n, D] = size(x); K = feval(cov{:}, hyp.cov, x); % evaluate covariance matrix m = feval(mean{:}, hyp.mean, x); % evaluate mean vector if numel(hyp.lik)==0 cov1 = cov{1}; if isa(cov1, 'function_handle'), cov1 = func2str(cov1); end if ~strcmp(cov1,'covSum'); error('Only covSum supported.'), end % check cov cov2 = cov{2}; npar = 0; for i=1:length(cov2) cov2i = cov2{i}; if numel(cov2i)==1 if isa(cov2i, 'function_handle'), cov2i = func2str(cov2i); end if strcmp(cov2i,'covNoise'), isn = npar+1; sn2 = exp(2*hyp.cov(isn)); end end npar = npar + eval(feval(cov2i)); end if ~exist('sn2','var'), error('We need covNoise in the covSum.'), end else % Gauss, Laplace, Sech2 sn2 = exp(2*hyp.lik(end)); isn = 0; K = K + sn2*eye(n); end L = chol(K)/sqrt(sn2); % Cholesky factor of covariance with noise alpha = solve_chol(L,y-m)/sn2; post.alpha = alpha; % return the posterior parameters post.sW = ones(n,1)/sqrt(sn2); % sqrt of noise precision vector post.L = L; % L = chol(eye(n)+sW*sW'.*K) if nargout>1 % do we want the marginal likelihood? sdeff = 1./post.sW; % effective Gaussian width mneff = K*alpha + m; % alpha = inv(K)*(mneff-m), effective Gaussian center % all loo marginal means and variances at once V = post.L'\diag(1./sdeff); r = sum(V.*V,1)'; % r=diag( inv(K) ) loo.fmu = mneff - alpha./r; % GPML book, eq. 5.12 loo.fs2 = 1./r - sn2; % GPML book, eq. 5.12 [loo.lp, loo.ymu, loo.ys2] = feval(lik,hyp.lik,y,loo.fmu,loo.fs2); nloo = -sum(loo.lp); % negative leave-one-out predictive probability if nargout>2 % do we want derivatives? [lZ,dlZmn,d2lZmn] = feval(lik,hyp.lik,y,loo.fmu,loo.fs2,'infEP'); dlZva = (d2lZmn+dlZmn.*dlZmn)/2; % derivative w.r.t. variance dnloo = hyp; iK = solve_chol(L,eye(n))/sn2; for j=1:numel(hyp.cov) dKj = feval(cov{:}, hyp.cov, x, [], j); Zj = solve_chol(L,dKj)/sn2; % GPML book, eq. 5.13 dva = sum(Zj.*iK,2)./(r.*r); % GPML book, eq. 5.13 dmn = (Zj*alpha)./r - alpha.*dva; % GPML book, eq. 5.13 if j==isn, dva = dva - 2*sn2; end % additional part for covNoise dnloo.cov(j) = -dlZmn'*dmn -dlZva'*dva; end for j=1:numel(hyp.lik) dnloo.lik(j) = -sum( feval(lik,hyp.lik,y,loo.fmu,loo.fs2,'infEP',j) ); if j==numel(hyp.lik) dva = 2*sn2*sum(iK.*iK,2)./(r.*r); % GPML book, eq. 5.13 dmn = 2*solve_chol(L,alpha)./r - alpha.*dva; % GPML book, eq. 5.13 dva = dva - 2*sn2; dnloo.lik(j) = dnloo.lik(j) -dlZmn'*dmn -dlZva'*dva; end end for j=1:numel(hyp.mean) dmj = feval(mean{:}, hyp.mean, x, j); dmn = solve_chol(L,dmj)./(sn2*r); dnloo.mean(j) = -dlZmn'*dmn; end end end gpml-matlab-v3.1-2010-09-27/util/binaryEPGP.m000644 000765 000024 00000005753 11422240325 020160 0ustar00carlstaff000000 000000 function varargout = binaryEPGP(hyper, covfunc, x, y, xstar) % binaryEPGP - The Expectation Propagation approximation for binary Gaussian % process classification. Two modes are possible: training or testing: if no % test cases are supplied, then the approximate negative log marginal % likelihood and its partial derivatives wrt the hyperparameters is computed; % this mode is used to fit the hyperparameters. If test cases are given, then % the test set predictive probabilities are returned. The program is flexible % in allowing a multitude of covariance functions. % % usage: [nlZ, dnlZ ] = binaryEPGP(hyper, covfunc, x, y); % or: [p, mu, s2, nlZ] = binaryEPGP(hyper, covfunc, x, y, xstar); % % where: % % hyper is a (column) vector of hyperparameters % covfunc is the name of the covariance function (see below) % lik is the name of the likelihood function (see below) % x is a n by D matrix of training inputs % y is a (column) vector (of size n) of binary +1/-1 targets % xstar is a nn by D matrix of test inputs % nlZ is the returned value of the negative log marginal likelihood % dnlZ is a (column) vector of partial derivatives of the negative % log marginal likelihood wrt each log hyperparameter % p is a (column) vector (of length nn) of predictive probabilities % mu is a (column) vector (of length nn) of predictive latent means % s2 is a (column) vector (of length nn) of predictive latent variances % % The length of the vector of hyperparameters depends on the covariance % function, as specified by the "covfunc" input to the function, specifying the % name of a covariance function. A number of different covariance function are % implemented, and it is not difficult to add new ones. See "help covFunctions" % for the details % % The function can conveniently be used with the "minimize" function to train % a Gaussian process, eg: % % [hyper, fX, i] = minimize(hyper, 'binaryEPGP', length, 'covSEiso', % 'logistic', x, y); % % Copyright (c) 2004, 2005, 2006, 2007 Carl Edward Rasmussen, 2010-07-21. if nargin<4 || nargin>5 disp('Usage: [nlZ, dnlZ ] = binaryEPGP(hyper, covfunc, x, y);') disp(' or: [p, mu, s2, nlZ] = binaryEPGP(hyper, covfunc, x, y, xstar);') return end % Note, this function is just a wrapper provided for backward compatibility, % the functionality is now provided by the more general gp function. lik = @likErf; inf = @infEP; mean = @meanZero; varargout = cell(nargout, 1); % allocate the right number of output arguments hyp.cov = hyper; if nargin==4 [varargout{:}] = gp(hyp,inf,mean,covfunc,lik,x,y); if nargout>1, varargout{2} = varargout{2}.cov; end else [ymu,ys2,fmu,fs2] = gp(hyp,inf,mean,covfunc,lik,x,y,xstar); if nargout>0, varargout{1} = (1+ymu)/2; end if nargout>1, varargout{2} = fmu; end if nargout>2, varargout{3} = fs2; end if nargout>3, varargout{4} = gp(hyp,inf,mean,covfunc,lik,x,y); end % nlZ end gpml-matlab-v3.1-2010-09-27/util/binaryLaplaceGP.m000644 000765 000024 00000007030 11422240325 021203 0ustar00carlstaff000000 000000 function varargout = binaryLaplaceGP(hyper, covfunc, lik, x, y, xstar) % binaryLaplaceGP - Laplace's approximation for binary Gaussian process % classification. Two modes are possible: training or testing: if no test % cases are supplied, then the approximate negative log marginal likelihood % and its partial derivatives wrt the hyperparameters is computed; this mode is % used to fit the hyperparameters. If test cases are given, then the test set % predictive probabilities are returned. The program is flexible in allowing % several different likelihood functions and a multitude of covariance % functions. % % usage: [nlZ, dnlZ ] = binaryLaplaceGP(hyper, covfunc, lik, x, y); % or: [p, mu, s2, nlZ] = binaryLaplaceGP(hyper, covfunc, lik, x, y, xstar); % % where: % % hyper is a (column) vector of hyperparameters % covfunc is the name of the covariance function (see below) % lik is the name of the likelihood function (see below) % x is a n by D matrix of training inputs % y is a (column) vector (of size n) of binary +1/-1 targets % xstar is a nn by D matrix of test inputs % nlZ is the returned value of the negative log marginal likelihood % dnlZ is a (column) vector of partial derivatives of the negative % log marginal likelihood wrt each log hyperparameter % p is a (column) vector (of length nn) of predictive probabilities % mu is a (column) vector (of length nn) of predictive latent means % s2 is a (column) vector (of length nn) of predictive latent variances % % The length of the vector of log hyperparameters depends on the covariance % function, as specified by the "covfunc" input to the function, specifying the % name of a covariance function. A number of different covariance function are % implemented, and it is not difficult to add new ones. See "help covFunctions" % for the details. % % The shape of the likelihood function is given by the "lik" input to the % function, specifying the name of the likelihood function. The two implemented % likelihood functions are: % % logistic the logistic function: 1/(1+exp(-x)) % cumGauss the cumulative Gaussian (error function) % % The function can conveniently be used with the "minimize" function to train % a Gaussian process, eg: % % [hyper, fX, i] = minimize(hyper, 'binaryLaplaceGP', length, 'covSEiso', % 'logistic', x, y); % % Copyright (c) 2004, 2005, 2006, 2007 Carl Edward Rasmussen, 2010-07-21. if nargin<5 || nargin>6 disp('Usage: [nlZ, dnlZ ] = binaryLaplaceGP(hyper, covfunc, lik, x, y);') disp(' or: [p, mu, s2, nlZ] = binaryLaplaceGP(hyper, covfunc, lik, x, y, xstar);') return end % Note, this function is just a wrapper provided for backward compatibility, % the functionality is now provided by the more general gp function. if strcmp(lik,'logistic') lik = @likLogistic; elseif strcmpi(lik,'cumgauss') lik = @likErf; else error('Allowable likelihoods: logistic and cumGauss.'); end inf = @infLaplace; mean = @meanZero; varargout = cell(nargout, 1); % allocate the right number of output arguments hyp.cov = hyper; if nargin==5 [varargout{:}] = gp(hyp,inf,mean,covfunc,lik,x,y); if nargout>1, varargout{2} = varargout{2}.cov; end else [ymu,ys2,fmu,fs2] = gp(hyp,inf,mean,covfunc,lik,x,y,xstar); if nargout>0, varargout{1} = (1+ymu)/2; end if nargout>1, varargout{2} = fmu; end if nargout>2, varargout{3} = fs2; end if nargout>3, varargout{4} = gp(hyp,inf,mean,covfunc,lik,x,y); end % nlZ end gpml-matlab-v3.1-2010-09-27/util/solve_chol.c000644 000765 000024 00000002404 11504370467 020345 0ustar00carlstaff000000 000000 /* solve_chol - solve a linear system A*X = B using the cholesky factorization of A (where A is square, symmetric and positive definite. Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-12-16. */ #include "mex.h" #include #include extern int dpotrs_(char *, long *, long *, double *, long *, double *, long *, long *); void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) { double *C; long n, m, q; if (nrhs != 2 || nlhs > 1) /* check the input */ mexErrMsgTxt("Usage: X = solve_chol(R, B)"); n = mxGetN(prhs[0]); if (n != mxGetM(prhs[0])) mexErrMsgTxt("Error: First argument matrix must be square"); if (n != mxGetM(prhs[1])) mexErrMsgTxt("Error: First and second argument matrices must have same number of rows"); m = mxGetN(prhs[1]); plhs[0] = mxCreateDoubleMatrix(n, m, mxREAL); /* allocate space for output */ C = mxGetPr(plhs[0]); if (n==0) return; /* if argument was empty matrix, do no more */ memcpy( C, mxGetPr(prhs[1]), n*m*sizeof(double) );/* copy argument matrix */ dpotrs_("U", &n, &m, mxGetPr(prhs[0]), &n, C, &n, &q); /* solve system */ if (q > 0) mexErrMsgTxt("Error: illegal input to solve_chol"); } gpml-matlab-v3.1-2010-09-27/util/solve_chol.m000644 000765 000024 00000001741 11447142675 020366 0ustar00carlstaff000000 000000 % solve_chol - solve linear equations from the Cholesky factorization. % Solve A*X = B for X, where A is square, symmetric, positive definite. The % input to the function is R the Cholesky decomposition of A and the matrix B. % Example: X = solve_chol(chol(A),B); % % NOTE: The program code is written in the C language for efficiency and is % contained in the file solve_chol.c, and should be compiled using matlabs mex % facility. However, this file also contains a (less efficient) matlab % implementation, supplied only as a help to people unfamiliar with mex. If % the C code has been properly compiled and is avaiable, it automatically % takes precendence over the matlab code in this file. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-09-18. function X = solve_chol(L, B) if nargin ~= 2 || nargout > 1 error('Wrong number of arguments.'); end if size(L,1) ~= size(L,2) || size(L,1) ~= size(B,1) error('Wrong sizes of matrix arguments.'); end X = L\(L'\B);gpml-matlab-v3.1-2010-09-27/util/lbfgsb.m000644 000765 000024 00000011601 11474502347 017460 0ustar00carlstaff000000 000000 % LBFGSB Call the nonlinear bound-constrained solver that uses % limited-memory BFGS quasi-Newton updates. % % The basic function call is % % LBFGSB(x0,lb,ub,objfunc,gradfunc) % % The first input argument x0 is either a matrix or a cell array of % matrices. It declares the starting point for the solver. % % The second and third input arguments lb and ub must be of the exact same % structure as x0. They declare the lower and upper bounds on the % variables, respectively. Set an entry to Inf to declare no bound. % % The next two input arguments must be the names of MATLAB routines % (M-files): % % objfunc Calculates the objective function at the current % point. The routine must accept as many inputs as cell % entry in x0 (or one input if x0 is a matrix). The % output must be a scalar representing the objective % evaluated at the current point. % % gradfunc Computes the gradient of the objective at the current % point. The input is the same as objfunc, but it must % return as many outputs as there are inputs, and each % of the outputs have the same matrix structure as its % corresponding input. % % Optionally, one may choose to pass additional auxiliary data to the % MATLAB callback routines listed above through the function call % LBFGSB(...,auxdata). If auxdata is the empty matrix, no extra % information is passed. It is important to observe that the auxiliary % data MAY NOT CHANGE through the course of the L-BFGS optimization! The % auxiliary data keep the same values as they possessed in the initial % call. If you need variables that change over time, you may want to % consider global variables (type HELP GLOBAL). % % LBFGSB(...,auxdata,iterfunc) specifies an additional callback routine % which is called once per algorithm iteration. The callback routine must % take the form ITERFUNC(T,F,X1,X2,...,AUXDATA). T is the current % iteration of the algorithm. F is the current value of the objective. The % inputs X1, X2,... are exactly the same as the ones passed to objfunc and % gradfunc. Finally, extra information may be passed through the input % AUXDATA. No outputs are expected from iterfunc. If iterfunc is the empty % string, no routine is called. % % These inputs may be followed by parameter/value pairs to modify the % default algorithm options. The algorithm option 'maxiter' sets the % maximum number of iterations. The rest of the options 'm', 'factr' and % 'pgtol' are described in the L-BFGS-B documentation. Note that in this % implementation we divide factr by the machine precision, so that the % convergence condition is satisfied when it is less than factr, not less % than factr times the machine precision. % % Interface has been written by Peter Carbonetto. % Compilation instructions by Hannes Nickisch, 2010-11-24 function x = lbfgsb(x,varargin) fprintf('\n\n\n') fprintf('##########################################################################\n') fprintf('In order to use the L-BFGS minimiser you have to compile Peter Carbonetto''s\n') fprintf('"Matlab interface for L-BFGS-B" [1]. The challenge here is the Fortran 77 code.\n') fprintf('We provide a Makefile suitable for Linux 32/64 bit and Mac whenever you have g77\n') fprintf('properly installed.Under Ubuntu, you can achieve this by installing the package\n') fprintf('fort77, for example.\n\n') fprintf('Compilation is done by editing lbfgsb/Makefile. In any case, you need to\n') fprintf('provide MATLAB_HOME which can be found by the commands ''locate matlab'' or \n') fprintf('''find / -name "matlab"''. You can choose between two compilation modes:\n') fprintf('a) using the mex utility by Matlab [default]\n') fprintf(' provide MEX, then type ''make mex''\n\n') fprintf('b) without mex utility by Matlab\n') fprintf(' provide MEX_SUFFIX and MATLAB_LIB, then type ''make nomex''\n\n') fprintf('In Ubuntu 10.04 LTS, the libg2c library needed for both a) and b) is not\n') fprintf('included per default. If ''ls /usr/lib/libg2c.*'' does not list anything\n') fprintf('this is the case on your machine. You then whant to install the packages\n') fprintf('gcc-3.4-base and libg2c0 e.g. from\n') fprintf('http://packages.ubuntu.com/hardy/gcc-3.4-base and\n') fprintf('http://packages.ubuntu.com/hardy/libg2c0.\n') fprintf('After installation, you have to create a symbolic link by ''cd /usr/lib'' and\n') fprintf('''ln -s libg2c.so.0 libg2c.so''.\n\n') fprintf('[1] http://www.cs.ubc.ca/~pcarbo/lbfgsb-for-matlab.html\n') fprintf('[2] http://www.mathworks.com/support/compilers/R2010a\n') fprintf('##########################################################################\n') fprintf('\n\n\n') error gpml-matlab-v3.1-2010-09-27/util/lbfgsb/000755 000765 000024 00000000000 11527514620 017277 5ustar00carlstaff000000 000000 gpml-matlab-v3.1-2010-09-27/util/minimize_lbfgsb_gradfun.m000644 000765 000024 00000004526 11422240325 023063 0ustar00carlstaff000000 000000 function G = minimize_lbfgsb_gradfun(X,varargin) % extract input arguments varargin = varargin{1}; strctX = varargin{2}; f = varargin{1}; % global variables serve as communication interface between calls global minimize_lbfgsb_iteration_number global minimize_lbfgsb_objective global minimize_lbfgsb_gradient global minimize_lbfgsb_X if norm(unwrap(X)-unwrap(minimize_lbfgsb_X))>1e-10 [y,G] = feval(f,rewrap(strctX,X),varargin{3:end}); else y = minimize_lbfgsb_objective(minimize_lbfgsb_iteration_number); G = minimize_lbfgsb_gradient; end % memorise gradient and position minimize_lbfgsb_gradient = G; minimize_lbfgsb_X = X; G = unwrap(G); % Extract the numerical values from "s" into the column vector "v". The % variable "s" can be of any type, including struct and cell array. % Non-numerical elements are ignored. See also the reverse rewrap.m. function v = unwrap(s) v = []; if isnumeric(s) v = s(:); % numeric values are recast to column vector elseif isstruct(s) v = unwrap(struct2cell(orderfields(s)));% alphabetize, conv to cell, recurse elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially v = [v; unwrap(s{i})]; end end % other types are ignored % Map the numerical elements in the vector "v" onto the variables "s" which can % be of any type. The number of numerical elements must match; on exit "v" % should be empty. Non-numerical entries are just copied. See also unwrap.m. function [s v] = rewrap(s, v) if isnumeric(s) if numel(v) < numel(s) error('The vector for conversion contains too few elements') end s = reshape(v(1:numel(s)), size(s)); % numeric values are reshaped v = v(numel(s)+1:end); % remaining arguments passed on elseif isstruct(s) [s p] = orderfields(s); p(p) = 1:numel(p); % alphabetize, store ordering [t v] = rewrap(struct2cell(s), v); % convert to cell, recurse s = orderfields(cell2struct(t,fieldnames(s),1),p); % conv to struct, reorder elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially [s{i} v] = rewrap(s{i}, v); end end % other types are not processed gpml-matlab-v3.1-2010-09-27/util/minimize_lbfgsb.m000644 000765 000024 00000010574 11422240325 021355 0ustar00carlstaff000000 000000 function [X, fX, i] = minimize_lbfgsb(X, f, length, varargin) % Minimize a differentiable multivariate function using quasi Newton. % % Usage: [X, fX, i] = minimize_lbfgsb(X, f, length, P1, P2, P3, ... ) % % X initial guess; may be of any type, including struct and cell array % f the name or pointer to the function to be minimized. The function % f must return two arguments, the value of the function, and it's % partial derivatives wrt the elements of X. The partial derivative % must have the same type as X. % length length of the run; if it is positive, it gives the maximum number of % line searches, if negative its absolute gives the maximum allowed % number of function evaluations. Optionally, length can have a second % component, which will indicate the reduction in function value to be % expected in the first line-search (defaults to 1.0). % P1, P2 ... parameters are passed to the function f. % % X the returned solution % fX vector of function values indicating progress made % i number of iterations (line searches or function evaluations, % depending on the sign of "length") used at termination. % % The function returns when either its length is up, or if no further progress % can be made (ie, we are at a (local) minimum, or so close that due to % numerical problems, we cannot get any closer). NOTE: If the function % terminates within a few iterations, it could be an indication that the % function values and derivatives are not consistent (ie, there may be a bug in % the implementation of your "f" function). % % Copyright (C) 2010 by Hannes Nickisch, 2010-02-05 % global variables serve as communication interface between calls global minimize_lbfgsb_iteration_number global minimize_lbfgsb_objective global minimize_lbfgsb_gradient global minimize_lbfgsb_X % init global variables minimize_lbfgsb_iteration_number = 0; minimize_lbfgsb_objective = Inf; minimize_lbfgsb_gradient = 0*unwrap(X); minimize_lbfgsb_X = X; X0 = X; lb = -Inf*ones(size(unwrap(X0))); ub = Inf*ones(size(unwrap(X0))); maxiter = abs(length); % max number of iterations % no callback routine used so far % m is the number of saved vectors used to estimate the Hessian % factr is the precision 1e-12 X = lbfgsb( unwrap(X0), lb, ub, 'minimize_lbfgsb_objfun', ... 'minimize_lbfgsb_gradfun', ... {f,X0,varargin{:}}, [], ... 'maxiter',maxiter, 'm',4, 'factr',1e-12, 'pgtol',1e-5); i = minimize_lbfgsb_iteration_number; fX = minimize_lbfgsb_objective; X = rewrap(X0,X); % clear global variables clear minimize_lbfgsb_iteration_number clear minimize_lbfgsb_objective clear minimize_lbfgsb_gradient clear minimize_lbfgsb_X % Extract the numerical values from "s" into the column vector "v". The % variable "s" can be of any type, including struct and cell array. % Non-numerical elements are ignored. See also the reverse rewrap.m. function v = unwrap(s) v = []; if isnumeric(s) v = s(:); % numeric values are recast to column vector elseif isstruct(s) v = unwrap(struct2cell(orderfields(s)));% alphabetize, conv to cell, recurse elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially v = [v; unwrap(s{i})]; end end % other types are ignored % Map the numerical elements in the vector "v" onto the variables "s" which can % be of any type. The number of numerical elements must match; on exit "v" % should be empty. Non-numerical entries are just copied. See also unwrap.m. function [s v] = rewrap(s, v) if isnumeric(s) if numel(v) < numel(s) error('The vector for conversion contains too few elements') end s = reshape(v(1:numel(s)), size(s)); % numeric values are reshaped v = v(numel(s)+1:end); % remaining arguments passed on elseif isstruct(s) [s p] = orderfields(s); p(p) = 1:numel(p); % alphabetize, store ordering [t v] = rewrap(struct2cell(s), v); % convert to cell, recurse s = orderfields(cell2struct(t,fieldnames(s),1),p); % conv to struct, reorder elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially [s{i} v] = rewrap(s{i}, v); end end % other types are not processed gpml-matlab-v3.1-2010-09-27/util/gpr.m000644 000765 000024 00000004667 11447142675 017033 0ustar00carlstaff000000 000000 function varargout = gpr(hyper, cov, x, y, xs) % gpr - Gaussian process regression, with a named covariance function. Two % modes are possible: training and prediction: if no test data are given, the % function returns minus the log likelihood and its partial derivatives with % respect to the hyperparameters; this mode is used to fit the hyperparameters. % If test data are given, then (marginal) Gaussian predictions are computed, % whose mean and variance are returned. Note that in cases where the covariance % function has noise contributions, the variance returned in s2 is for noisy % test targets; if you want the variance of the noise-free latent function, you % must substract the noise variance. % % usage: [nlZ dnlZ] = gpr(hyp, cov, x, y) % or: [mu s2] = gpr(hyp, cov, x, y, xs) % % where: % % hyp is a (column) vector of log hyperparameters % cov is the covariance function % x is a n by D matrix of training inputs % y is a (column) vector (of size n) of targets % xs is a ns by D matrix of test inputs % nlZ is the returned value of the negative log marginal likelihood % dnlZ is a (column) vector of partial derivatives of the negative % log marginal likelihood wrt each log hyperparameter % mu is a (column) vector (of size nn) of prediced means % s2 is a (column) vector (of size nn) of predicted variances % % For more help on covariance functions, see "help covFunctions". % % Copyright (c) 2010 Carl Edward Rasmussen and Hannes Nickisch 2010-06-18. err = 'we need cov = {''covSum'', {cov1, ..,''covNoise'', .., covP} }; to map to gp.m'; if ~strcmp(cov{1},'covSum'), error(err), end id = 0; nhyp = zeros(length(cov{2}),1); for i=1:length(cov{2}) if strcmp(cov{2}{i},'covNoise'), id = i; break, end nhyp(i) = eval(feval(cov{2}{i})); end if id==0, error(err), else nhyp = sum(nhyp(1:id-1)); end cov{2} = cov{2}([1:id-1,id+1:end]); hyp.lik = hyper(nhyp+1); hyp.cov = hyper([1:nhyp,nhyp+2:end]); % Note, this function is just a wrapper provided for backward compatibility, % the functionality is now provided by the more general gp function. lik = @likGauss; inf = @infExact; mean = @meanZero; varargout = cell(nargout, 1); % allocate the right number of output arguments if nargin==4 [varargout{:}] = gp(hyp,inf,mean,cov,lik,x,y); if nargout>1, varargout{2} = [varargout{2}.lik; varargout{2}.cov]; end else [varargout{:}] = gp(hyp,inf,mean,cov,lik,x,y,xs); end gpml-matlab-v3.1-2010-09-27/util/sq_dist.m000644 000765 000024 00000003657 11504370467 017703 0ustar00carlstaff000000 000000 % sq_dist - a function to compute a matrix of all pairwise squared distances % between two sets of vectors, stored in the columns of the two matrices, a % (of size D by n) and b (of size D by m). If only a single argument is given % or the second matrix is empty, the missing matrix is taken to be identical % to the first. % % Usage: C = sq_dist(a, b) % or: C = sq_dist(a) or equiv.: C = sq_dist(a, []) % % Where a is of size Dxn, b is of size Dxm (or empty), C is of size nxm. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-12-13. function C = sq_dist(a, b) if nargin<1 || nargin>3 || nargout>1, error('Wrong number of arguments.'); end bsx = exist('bsxfun','builtin'); % since Matlab R2007a 7.4.0 and Octave 3.0 if ~bsx, bsx = exist('bsxfun'); end % bsxfun is not yes "builtin" in Octave [D, n] = size(a); % Computation of a^2 - 2*a*b + b^2 is less stable than (a-b)^2 because numerical % precision can be lost when both a and b have very large absolute value and the % same sign. For that reason, we subtract the mean from the data beforehand to % stabilise the computations. This is OK because the squared error is % independent of the mean. if nargin==1 % subtract mean mu = mean(a,2); if bsx a = bsxfun(@minus,a,mu); else a = a - repmat(mu,1,size(a,2)); end b = a; m = n; else [d, m] = size(b); if d ~= D, error('Error: column lengths must agree.'); end mu = (m/(n+m))*mean(b,2) + (n/(n+m))*mean(a,2); if bsx a = bsxfun(@minus,a,mu); b = bsxfun(@minus,b,mu); else a = a - repmat(mu,1,n); b = b - repmat(mu,1,m); end end if bsx % compute squared distances C = bsxfun(@plus,sum(a.*a,1)',bsxfun(@minus,sum(b.*b,1),2*a'*b)); else C = repmat(sum(a.*a,1)',1,m) + repmat(sum(b.*b,1),n,1) - 2*a'*b; end C = max(C,0); % numerical noise can cause C to negative i.e. C > -1e-14 gpml-matlab-v3.1-2010-09-27/util/make.m000644 000765 000024 00000004064 11504370467 017143 0ustar00carlstaff000000 000000 % For compilation under Linux/Octave you need to install the package % octavex.x-headers. % Under Windows/Octave, compilation works out of the box. % % There is a big variety of platforms running Matlab around i.e. % 'SUN4','SOL2','HP700','SGI','SGI64','IBM_RS','ALPHA', % 'LNX86','GLNX86','GLNXA64', % 'MAC', 'MACI' % the following script is only tested for the most common ones (Mac,Linux,Win) % but should work with minor modifications on your architecture. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch 2010-12-16. OCTAVE = exist('OCTAVE_VERSION') ~= 0; % check if we run Matlab or Octave %% 0) change directory me = mfilename; % what is my filename mydir = which(me); mydir = mydir(1:end-2-numel(me)); % where am I located cd(mydir) if OCTAVE %% compile solve_chol mkoctfile --mex solve_chol.c delete solve_chol.o else % MATLAB %% compile solve_chol if ispc % Windows libext = 'lib'; % remove the trailing underscore after the Lapack function "dpotrs" % we do that in a clumsy way since there is no sed/awk under Windows fid = fopen('solve_chol.c','r'); fid_win = fopen('solve_chol2XXXX.c','w'); while 1 tline = fgetl(fid); if ~ischar(tline), break, end id = strfind(tline,'dpotrs'); if numel(id) tline = [tline(1:id-1),'dpotrs',tline(id+7:end)]; end fprintf(fid_win,[tline,'\n']); end fclose(fid_win); fclose(fid); if numel(strfind(computer,'64')) % 64 bit ospath = 'extern/lib/win64/microsoft'; else % 32 bit ospath = 'extern/lib/win32/lcc'; end % compile comp_cmd = 'mex -O solve_chol2XXXX.c -output solve_chol'; eval([comp_cmd,' ''',matlabroot,'/',ospath,'/libmwlapack.',libext,'''']) delete solve_chol2XXXX* else % Linux, MAC, Solaris mex -O -lmwlapack solve_chol.c end endgpml-matlab-v3.1-2010-09-27/util/unwrap.m000644 000765 000024 00000001213 11422240325 017517 0ustar00carlstaff000000 000000 % Extract the numerical values from "s" into the column vector "v". The % variable "s" can be of any type, including struct and cell array. % Non-numerical elements are ignored. See also the reverse rewrap.m. function v = unwrap(s) v = []; if isnumeric(s) v = s(:); % numeric values are recast to column vector elseif isstruct(s) v = unwrap(struct2cell(orderfields(s))); % alphabetize, conv to cell, recurse elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially v = [v; unwrap(s{i})]; end end % other types are ignored gpml-matlab-v3.1-2010-09-27/util/brentmin.m000644 000765 000024 00000007574 11422240325 020041 0ustar00carlstaff000000 000000 %% BRENTMIN: Brent's minimization method in one dimension function [xmin,fmin,funccount,varargout] = ... brentmin(xlow,xupp,Nitmax,tol,f,nout,varargin) % code taken from % § 10.2 Parabolic Interpolation and Brent's Method in One Dimension % Press, Teukolsky, Vetterling & Flannery % Numerical Recipes in C, Cambridge University Press, 2002 % % [xmin,fmin,funccout,varargout] = BRENTMIN(xlow,xupp,Nit,tol,f,nout,varargin) % Given a function f, and given a search interval this routine isolates % the minimum of fractional precision of about tol using Brent's method. % % INPUT % ----- % xlow,xupp: search interval such that xlow<=xmin<=xupp % Nitmax: maximum number of function evaluations made by the routine % tol: fractional precision % f: [y,varargout{:}] = f(x,varargin{:}) is the function % nout: no. of outputs of f (in varargout) in addition to the y value % % OUTPUT % ------ % fmin: minimal function value % xmin: corresponding abscissa-value % funccount: number of function evaluations made % varargout: additional outputs of f at optimum % % Copyright (c) by Hannes Nickisch 2010-01-10. if nargin<6, nout = 0; end varargout = cell(nout,1); % tolerance is no smaller than machine's floating point precision tol = max(tol,eps); % Evaluate endpoints fa = f(xlow,varargin{:}); fb = f(xupp,varargin{:}); funccount = 2; % number of function evaluations % Compute the start point seps = sqrt(eps); c = 0.5*(3.0 - sqrt(5.0));% golden ratio a = xlow; b = xupp; v = a + c*(b-a); w = v; xf = v; d = 0.0; e = 0.0; x = xf; [fx,varargout{:}] = f(x,varargin{:}); funccount = funccount + 1; fv = fx; fw = fx; xm = 0.5*(a+b); tol1 = seps*abs(xf) + tol/3.0; tol2 = 2.0*tol1; % Main loop while ( abs(xf-xm) > (tol2 - 0.5*(b-a)) ) gs = 1; % Is a parabolic fit possible if abs(e) > tol1 % Yes, so fit parabola gs = 0; r = (xf-w)*(fx-fv); q = (xf-v)*(fx-fw); p = (xf-v)*q-(xf-w)*r; q = 2.0*(q-r); if q > 0.0, p = -p; end q = abs(q); r = e; e = d; % Is the parabola acceptable if ( (abs(p)q*(a-xf)) && (p= xm, e = a-xf; else e = b-xf; end d = c*e; end % The function must not be evaluated too close to xf si = sign(d) + (d == 0); x = xf + si * max( abs(d), tol1 ); [fu,varargout{:}] = f(x,varargin{:}); funccount = funccount + 1; % Update a, b, v, w, x, xm, tol1, tol2 if fu <= fx if x >= xf, a = xf; else b = xf; end v = w; fv = fw; w = xf; fw = fx; xf = x; fx = fu; else % fu > fx if x < xf, a = x; else b = x; end if ( (fu <= fw) || (w == xf) ) v = w; fv = fw; w = x; fw = fu; elseif ( (fu <= fv) || (v == xf) || (v == w) ) v = x; fv = fu; end end xm = 0.5*(a+b); tol1 = seps*abs(xf) + tol/3.0; tol2 = 2.0*tol1; if funccount >= Nitmax % typically we should not get here % warning(sprintf(['Maximum number of iterations (%d) exceeded:', ... % 'precision is not guaranteed'],Nitmax)) % fprintf('[%1.3f,%1.3f,%1.3f]\n',xlow,xf,xupp) break end end % while % check that endpoints are less than the minimum found if ( (fa < fx) && (fa <= fb) ) xf = xlow; fx = fa; elseif fb < fx xf = xupp; fx = fb; end fmin = fx; xmin = xf; gpml-matlab-v3.1-2010-09-27/util/minimize.m000644 000765 000024 00000025676 11422240325 020047 0ustar00carlstaff000000 000000 function [X, fX, i] = minimize(X, f, length, varargin) % Minimize a differentiable multivariate function using conjugate gradients. % % Usage: [X, fX, i] = minimize(X, f, length, P1, P2, P3, ... ) % % X initial guess; may be of any type, including struct and cell array % f the name or pointer to the function to be minimized. The function % f must return two arguments, the value of the function, and it's % partial derivatives wrt the elements of X. The partial derivative % must have the same type as X. % length length of the run; if it is positive, it gives the maximum number of % line searches, if negative its absolute gives the maximum allowed % number of function evaluations. Optionally, length can have a second % component, which will indicate the reduction in function value to be % expected in the first line-search (defaults to 1.0). % P1, P2, ... parameters are passed to the function f. % % X the returned solution % fX vector of function values indicating progress made % i number of iterations (line searches or function evaluations, % depending on the sign of "length") used at termination. % % The function returns when either its length is up, or if no further progress % can be made (ie, we are at a (local) minimum, or so close that due to % numerical problems, we cannot get any closer). NOTE: If the function % terminates within a few iterations, it could be an indication that the % function values and derivatives are not consistent (ie, there may be a bug in % the implementation of your "f" function). % % The Polack-Ribiere flavour of conjugate gradients is used to compute search % directions, and a line search using quadratic and cubic polynomial % approximations and the Wolfe-Powell stopping criteria is used together with % the slope ratio method for guessing initial step sizes. Additionally a bunch % of checks are made to make sure that exploration is taking place and that % extrapolation will not be unboundedly large. % % See also: checkgrad % % Copyright (C) 2001 - 2010 by Carl Edward Rasmussen, 2010-01-03 INT = 0.1; % don't reevaluate within 0.1 of the limit of the current bracket EXT = 3.0; % extrapolate maximum 3 times the current step-size MAX = 20; % max 20 function evaluations per line search RATIO = 10; % maximum allowed slope ratio SIG = 0.1; RHO = SIG/2; % SIG and RHO are the constants controlling the Wolfe- % Powell conditions. SIG is the maximum allowed absolute ratio between % previous and new slopes (derivatives in the search direction), thus setting % SIG to low (positive) values forces higher precision in the line-searches. % RHO is the minimum allowed fraction of the expected (from the slope at the % initial point in the linesearch). Constants must satisfy 0 < RHO < SIG < 1. % Tuning of SIG (depending on the nature of the function to be optimized) may % speed up the minimization; it is probably not worth playing much with RHO. % The code falls naturally into 3 parts, after the initial line search is % started in the direction of steepest descent. 1) we first enter a while loop % which uses point 1 (p1) and (p2) to compute an extrapolation (p3), until we % have extrapolated far enough (Wolfe-Powell conditions). 2) if necessary, we % enter the second loop which takes p2, p3 and p4 chooses the subinterval % containing a (local) minimum, and interpolates it, unil an acceptable point % is found (Wolfe-Powell conditions). Note, that points are always maintained % in order p0 <= p1 <= p2 < p3 < p4. 3) compute a new search direction using % conjugate gradients (Polack-Ribiere flavour), or revert to steepest if there % was a problem in the previous line-search. Return the best value so far, if % two consecutive line-searches fail, or whenever we run out of function % evaluations or line-searches. During extrapolation, the "f" function may fail % either with an error or returning Nan or Inf, and minimize should handle this % gracefully. if max(size(length)) == 2, red=length(2); length=length(1); else red=1; end if length>0, S='Linesearch'; else S='Function evaluation'; end i = 0; % zero the run length counter ls_failed = 0; % no previous line search has failed [f0 df0] = feval(f, X, varargin{:}); % get function value and gradient Z = X; X = unwrap(X); df0 = unwrap(df0); fprintf('%s %6i; Value %4.6e\r', S, i, f0); if exist('fflush','builtin') fflush(stdout); end fX = f0; i = i + (length<0); % count epochs?! s = -df0; d0 = -s'*s; % initial search direction (steepest) and slope x3 = red/(1-d0); % initial step is red/(|s|+1) while i < abs(length) % while not finished i = i + (length>0); % count iterations?! X0 = X; F0 = f0; dF0 = df0; % make a copy of current values if length>0, M = MAX; else M = min(MAX, -length-i); end while 1 % keep extrapolating as long as necessary x2 = 0; f2 = f0; d2 = d0; f3 = f0; df3 = df0; success = 0; while ~success && M > 0 try M = M - 1; i = i + (length<0); % count epochs?! [f3 df3] = feval(f, rewrap(Z,X+x3*s), varargin{:}); df3 = unwrap(df3); if isnan(f3) || isinf(f3) || any(isnan(df3)+isinf(df3)), error(''), end success = 1; catch % catch any error which occured in f x3 = (x2+x3)/2; % bisect and try again end end if f3 < F0, X0 = X+x3*s; F0 = f3; dF0 = df3; end % keep best values d3 = df3'*s; % new slope if d3 > SIG*d0 || f3 > f0+x3*RHO*d0 || M == 0 % are we done extrapolating? break end x1 = x2; f1 = f2; d1 = d2; % move point 2 to point 1 x2 = x3; f2 = f3; d2 = d3; % move point 3 to point 2 A = 6*(f1-f2)+3*(d2+d1)*(x2-x1); % make cubic extrapolation B = 3*(f2-f1)-(2*d1+d2)*(x2-x1); x3 = x1-d1*(x2-x1)^2/(B+sqrt(B*B-A*d1*(x2-x1))); % num. error possible, ok! if ~isreal(x3) || isnan(x3) || isinf(x3) || x3 < 0 % num prob | wrong sign? x3 = x2*EXT; % extrapolate maximum amount elseif x3 > x2*EXT % new point beyond extrapolation limit? x3 = x2*EXT; % extrapolate maximum amount elseif x3 < x2+INT*(x2-x1) % new point too close to previous point? x3 = x2+INT*(x2-x1); end end % end extrapolation while (abs(d3) > -SIG*d0 || f3 > f0+x3*RHO*d0) && M > 0 % keep interpolating if d3 > 0 || f3 > f0+x3*RHO*d0 % choose subinterval x4 = x3; f4 = f3; d4 = d3; % move point 3 to point 4 else x2 = x3; f2 = f3; d2 = d3; % move point 3 to point 2 end if f4 > f0 x3 = x2-(0.5*d2*(x4-x2)^2)/(f4-f2-d2*(x4-x2)); % quadratic interpolation else A = 6*(f2-f4)/(x4-x2)+3*(d4+d2); % cubic interpolation B = 3*(f4-f2)-(2*d2+d4)*(x4-x2); x3 = x2+(sqrt(B*B-A*d2*(x4-x2)^2)-B)/A; % num. error possible, ok! end if isnan(x3) || isinf(x3) x3 = (x2+x4)/2; % if we had a numerical problem then bisect end x3 = max(min(x3, x4-INT*(x4-x2)),x2+INT*(x4-x2)); % don't accept too close [f3 df3] = feval(f, rewrap(Z,X+x3*s), varargin{:}); df3 = unwrap(df3); if f3 < F0, X0 = X+x3*s; F0 = f3; dF0 = df3; end % keep best values M = M - 1; i = i + (length<0); % count epochs?! d3 = df3'*s; % new slope end % end interpolation if abs(d3) < -SIG*d0 && f3 < f0+x3*RHO*d0 % if line search succeeded X = X+x3*s; f0 = f3; fX = [fX' f0]'; % update variables fprintf('%s %6i; Value %4.6e\r', S, i, f0); if exist('fflush','builtin') fflush(stdout); end s = (df3'*df3-df0'*df3)/(df0'*df0)*s - df3; % Polack-Ribiere CG direction df0 = df3; % swap derivatives d3 = d0; d0 = df0'*s; if d0 > 0 % new slope must be negative s = -df0; d0 = -s'*s; % otherwise use steepest direction end x3 = x3 * min(RATIO, d3/(d0-realmin)); % slope ratio but max RATIO ls_failed = 0; % this line search did not fail else X = X0; f0 = F0; df0 = dF0; % restore best point so far if ls_failed || i > abs(length) % line search failed twice in a row break; % or we ran out of time, so we give up end s = -df0; d0 = -s'*s; % try steepest x3 = 1/(1-d0); ls_failed = 1; % this line search failed end end X = rewrap(Z,X); fprintf('\n'); if exist('fflush','builtin') fflush(stdout); end % Extract the numerical values from "s" into the column vector "v". The % variable "s" can be of any type, including struct and cell array. % Non-numerical elements are ignored. See also the reverse rewrap.m. function v = unwrap(s) v = []; if isnumeric(s) v = s(:); % numeric values are recast to column vector elseif isstruct(s) v = unwrap(struct2cell(orderfields(s))); % alphabetize, conv to cell, recurse elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially v = [v; unwrap(s{i})]; end end % other types are ignored % Map the numerical elements in the vector "v" onto the variables "s" which can % be of any type. The number of numerical elements must match; on exit "v" % should be empty. Non-numerical entries are just copied. See also unwrap.m. function [s v] = rewrap(s, v) if isnumeric(s) if numel(v) < numel(s) error('The vector for conversion contains too few elements') end s = reshape(v(1:numel(s)), size(s)); % numeric values are reshaped v = v(numel(s)+1:end); % remaining arguments passed on elseif isstruct(s) [s p] = orderfields(s); p(p) = 1:numel(p); % alphabetize, store ordering [t v] = rewrap(struct2cell(s), v); % convert to cell, recurse s = orderfields(cell2struct(t,fieldnames(s),1),p); % conv to struct, reorder elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially [s{i} v] = rewrap(s{i}, v); end end % other types are not processed gpml-matlab-v3.1-2010-09-27/util/minimize_lbfgsb_objfun.m000644 000765 000024 00000005207 11422240325 022715 0ustar00carlstaff000000 000000 function y = minimize_lbfgsb_objfun(X,varargin) % extract input arguments varargin = varargin{1}; strctX = varargin{2}; f = varargin{1}; % global variables serve as communication interface between calls global minimize_lbfgsb_iteration_number global minimize_lbfgsb_objective global minimize_lbfgsb_gradient global minimize_lbfgsb_X recompute = 0; if minimize_lbfgsb_iteration_number==0 recompute = 1; else if norm(unwrap(X)-unwrap(minimize_lbfgsb_X))>1e-10 recompute = 1; end end if recompute [y,G] = feval(f,rewrap(strctX,X),varargin{3:end}); else y = minimize_lbfgsb_objective(minimize_lbfgsb_iteration_number); G = minimize_lbfgsb_gradient; end % increase global counter, memorise objective function, gradient and position minimize_lbfgsb_iteration_number = minimize_lbfgsb_iteration_number+1; minimize_lbfgsb_objective(minimize_lbfgsb_iteration_number,1) = y; minimize_lbfgsb_gradient = G; minimize_lbfgsb_X = X; % Extract the numerical values from "s" into the column vector "v". The % variable "s" can be of any type, including struct and cell array. % Non-numerical elements are ignored. See also the reverse rewrap.m. function v = unwrap(s) v = []; if isnumeric(s) v = s(:); % numeric values are recast to column vector elseif isstruct(s) v = unwrap(struct2cell(orderfields(s)));% alphabetize, conv to cell, recurse elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially v = [v; unwrap(s{i})]; end end % other types are ignored % Map the numerical elements in the vector "v" onto the variables "s" which can % be of any type. The number of numerical elements must match; on exit "v" % should be empty. Non-numerical entries are just copied. See also unwrap.m. function [s v] = rewrap(s, v) if isnumeric(s) if numel(v) < numel(s) error('The vector for conversion contains too few elements') end s = reshape(v(1:numel(s)), size(s)); % numeric values are reshaped v = v(numel(s)+1:end); % remaining arguments passed on elseif isstruct(s) [s p] = orderfields(s); p(p) = 1:numel(p); % alphabetize, store ordering [t v] = rewrap(struct2cell(s), v); % convert to cell, recurse s = orderfields(cell2struct(t,fieldnames(s),1),p); % conv to struct, reorder elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially [s{i} v] = rewrap(s{i}, v); end end % other types are not processed gpml-matlab-v3.1-2010-09-27/util/rewrap.m000644 000765 000024 00000001766 11422240325 017520 0ustar00carlstaff000000 000000 % Map the numerical elements in the vector "v" onto the variables "s" which can % be of any type. The number of numerical elements must match; on exit "v" % should be empty. Non-numerical entries are just copied. See also unwrap.m. function [s v] = rewrap(s, v) if isnumeric(s) if numel(v) < numel(s) error('The vector for conversion contains too few elements') end s = reshape(v(1:numel(s)), size(s)); % numeric values are reshaped v = v(numel(s)+1:end); % remaining arguments passed on elseif isstruct(s) [s p] = orderfields(s); p(p) = 1:numel(p); % alphabetize, store ordering [t v] = rewrap(struct2cell(s), v); % convert to cell, recurse s = orderfields(cell2struct(t,fieldnames(s),1),p); % conv to struct, reorder elseif iscell(s) for i = 1:numel(s) % cell array elements are handled sequentially [s{i} v] = rewrap(s{i}, v); end end % other types are not processed gpml-matlab-v3.1-2010-09-27/util/binaryGP.m000644 000765 000024 00000011547 11422240325 017731 0ustar00carlstaff000000 000000 function varargout = binaryGP(hyper, approx, covfunc, lik, x, y, xstar) % Approximate binary Gaussian Process classification. Two modes are possible: % training or testing: if no test cases are supplied, then the approximate % negative log marginal likelihood and its partial derivatives wrt the % hyperparameters is computed; this mode is used to fit the hyperparameters. If % test cases are given, then the test set predictive probabilities are % returned. Exact inference is intractible, the function uses a specified % approximation method (see approximations.m), flexible covariance functions % (see covFunctions.m) and likelihood functions (see likelihoods.m). % % usage: [nlZ, dnlZ ] = binaryGP(hyper, approx, covfunc, lik, x, y); % or: [p,mu,s2,nlZ] = binaryGP(hyper, approx, covfunc, lik, x, y, xstar); % % where: % % hyper is a column vector of hyperparameters % approx is a function specifying an approximation method for inference % covfunc is the name of the covariance function (see below) % lik is the name of the likelihood function % x is a n by D matrix of training inputs % y is a (column) vector (of size n) of binary +1/-1 targets % xstar is a nn by D matrix of test inputs % nlZ is the returned value of the negative log marginal likelihood % dnlZ is a (column) vector of partial derivatives of the negative % log marginal likelihood wrt each hyperparameter % p is a (column) vector (of length nn) of predictive probabilities % mu is a (column) vector (of length nn) of predictive latent means % s2 is a (column) vector (of length nn) of predictive latent variances % % The length of the vector of hyperparameters depends on the covariance % function, as specified by the "covfunc" input to the function, specifying the % name of a covariance function. A number of different covariance function are % implemented, and it is not difficult to add new ones. See covFunctions.m for % the details. % % The "lik" input argument specifies the name of the likelihood function (see % likelihoods.m). % % The "approx" input argument to the function specifies an approximation method % (see approximations.m). An approximation method returns a representation of % the approximate Gaussian posterior. Usually, the approximate posterior admits % the form N(m=K*alpha, V=inv(inv(K)+W)), where alpha is a vector and W is % diagonal. The approximation method returns: % % alpha is a (sparse or full column vector) containing inv(K)*m, where K % is the prior covariance matrix and m the approx posterior mean % sW is a (sparse or full column) vector containing diagonal of sqrt(W) % the approximate posterior covariance matrix is inv(inv(K)+W) % L is a (sparse or full) matrix, L = chol(sW*K*sW+eye(n)) % % In cases where the approximate posterior variance does not admit the form % V=inv(inv(K)+W) with diagonal W, L contains instead -inv(K+inv(W)), and sW % is unused. % % The alpha parameter may be sparse. In that case sW and L can either be sparse % or full (retaining only the non-zero rows and columns, as indicated by the % sparsity structure of alpha). The L paramter is allowed to be empty, in % which case it will be computed. % % The function can conveniently be used with the "minimize" function to train % a Gaussian Process, eg: % % [hyper, fX, i] = minimize(hyper, 'binaryGP', length, 'approxEP', 'covSEiso', 'logistic', x, y); % % where "length" gives the length of the run: if it is positive, it gives the % maximum number of line searches, if negative its absolute gives the maximum % allowed number of function evaluations. % % Copyright (c) 2007 Carl Edward Rasmussen and Hannes Nickisch, 2010-07-21. if nargin<6 || nargin>7 disp('Usage: [nlZ, dnlZ ] = binaryGP(hyper,approx,covfunc,lik,x,y);') disp(' or: [p,mu,s2,nlZ] = binaryGP(hyper,approx,covfunc,lik,x,y,xstar);') return end % Note, this function is just a wrapper provided for backward compatibility, % the functionality is now provided by the more general gp function. if strcmp(lik,'logistic') lik = @likLogistic; elseif strcmpi(lik,'cumgauss') lik = @likErf; else error('Allowable likelihoods: logistic and cumGauss.'); end if strcmp(approx,'approxLA') inf = @infLaplace; elseif strcmp(approx,'approxEP') inf = @infEP; else error('Allowable approximations: approxLA and approxEP.'); end mean = @meanZero; varargout = cell(nargout, 1); % allocate the right number of output arguments hyp.cov = hyper; if nargin==6 [varargout{:}] = gp(hyp,inf,mean,covfunc,lik,x,y); if nargout>1, varargout{2} = varargout{2}.cov; end else [ymu,ys2,fmu,fs2] = gp(hyp,inf,mean,covfunc,lik,x,y,xstar); if nargout>0, varargout{1} = (1+ymu)/2; end if nargout>1, varargout{2} = fmu; end if nargout>2, varargout{3} = fs2; end if nargout>3, varargout{4} = gp(hyp,inf,mean,covfunc,lik,x,y); end % nlZ endgpml-matlab-v3.1-2010-09-27/util/elsympol.m000644 000765 000024 00000001273 11422240325 020055 0ustar00carlstaff000000 000000 % Evaluate the order R elementary symmetric polynomial Newton's identity aka % the Newton–Girard formulae: http://en.wikipedia.org/wiki/Newton's_identities % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-01-10. function E = elsympol(Z,R) % evaluate 'power sums' of the individual terms in Z sz = size(Z); P = zeros([sz(1:2),R]); for r=1:R, P(:,:,r) = sum(Z.^r,3); end E = zeros([sz(1:2),R+1]); % E(:,:,r+1) yields polynomial r E(:,:,1) = ones(sz(1:2)); if R==0, return, end % init recursion E(:,:,2) = P(:,:,1); if R==1, return, end % init recursion for r=2:R for i=1:r E(:,:,r+1) = E(:,:,r+1) + P(:,:,i).*E(:,:,r+1-i)*(-1)^(i-1)/r; end end gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabscalar.cpp000644 000765 000024 00000001730 11422240325 022422 0ustar00carlstaff000000 000000 #include "matlabscalar.h" #include "matlabexception.h" // Function definitions. // ----------------------------------------------------------------- double& getMatlabScalar (const mxArray* ptr) { if (!mxIsDouble(ptr)) throw MatlabException("Matlab array must be of type double"); if (mxGetNumberOfElements(ptr) != 1) throw MatlabException("The Matlab array must be a scalar"); return *mxGetPr(ptr); } double& createMatlabScalar (mxArray*& ptr) { ptr = mxCreateDoubleScalar(0); return *mxGetPr(ptr); } // Function definitions for class MatlabScalar. // --------------------------------------------------------------- MatlabScalar::MatlabScalar (const mxArray* ptr) : x(getMatlabScalar(ptr)) { } MatlabScalar::MatlabScalar (mxArray*& ptr, double value) : x(createMatlabScalar(ptr)) { x = value; } MatlabScalar::MatlabScalar (MatlabScalar& source) : x(source.x) { } MatlabScalar& MatlabScalar::operator= (double value) { x = value; return *this; } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/program.cpp000644 000765 000024 00000007675 11422240325 021461 0ustar00carlstaff000000 000000 #include "program.h" #include // Function definitions. // ----------------------------------------------------------------- // Copy a C-style string (a null-terminated character array) to a // non-C-style string (a simple character array). The length of the // destination character array is given by "ndest". If the source is // shorter than the destination, the destination is padded with blank // characters. void copyCStrToCharArray (const char* source, char* dest, int ndest) { // Get the length of the source C string. int nsource = strlen(source); // Only perform the copy if the source can fit into the destination. if (nsource < ndest) { // Copy the string. strcpy(dest,source); // Fill in the rest of the string with blanks. for (int i = nsource; i < ndest; i++) dest[i] = ' '; } } // Return true if the two strings are the same. The second input // argument must be a C-style string (a null-terminated character // array), but the first input argument need not be one. We only // compare the two strings up to the length of "cstr". bool strIsEqualToCStr (const char* str, const char* cstr) { // Get the length of the C string. int n = strlen(cstr); return !strncmp(str,cstr,n); } // Function definitions for class Program. // ----------------------------------------------------------------- Program::Program (int n, double* x, double* lb, double* ub, int* btype, int m, int maxiter, double factr, double pgtol) { this->n = n; this->x = x; this->lb = lb; this->ub = ub; this->btype = btype; this->m = m; this->maxiter = maxiter; this->factr = factr; this->pgtol = pgtol; this->iprint = defaultprintlevel; this->owner = false; initStructures(); } Program::Program (int n, int m, int maxiter, double factr, double pgtol) { this->n = n; this->x = new double[n]; this->lb = new double[n]; this->ub = new double[n]; this->btype = new int[n]; this->m = m; this->maxiter = maxiter; this->factr = factr; this->pgtol = pgtol; this->iprint = defaultprintlevel; this->owner = true; initStructures(); } Program::~Program() { if (owner) { delete[] x; delete[] lb; delete[] ub; delete[] btype; } delete[] g; delete[] wa; delete[] iwa; } SolverExitStatus Program::runSolver() { SolverExitStatus status = success; // The return value. // Initialize the objective function and gradient to zero. f = 0; for (int i = 0; i < n; i++) g[i] = 0; // This initial call sets up the structures for L-BFGS. callLBFGS("START"); // Repeat until we've reached the maximum number of iterations. int t = 0; while (true) { // Do something according to the "task" from the previous call to // L-BFGS. if (strIsEqualToCStr(task,"FG")) { // Evaluate the objective function and the gradient of the // objective at the current point. f = computeObjective(n,x); computeGradient(n,x,g); } else if (strIsEqualToCStr(task,"NEW_X")) { // Go to the next iteration and call the iterative callback // routine. t++; iterCallback(t,x,f); // If we've reached the maximum number of iterations, terminate // the optimization. if (t == maxiter) { callLBFGS("STOP"); break; } } else if (strIsEqualToCStr(task,"CONV")) break; else if (strIsEqualToCStr(task,"ABNO")) { status = abnormalTermination; break; } else if (strIsEqualToCStr(task,"ERROR")) { status = errorOnInput; break; } // Call L-BFGS again. callLBFGS(); } return status; } void Program::callLBFGS (const char* cmd) { if (cmd) copyCStrToCharArray(cmd,task,60); setulb_(&n,&m,x,lb,ub,btype,&f,g,&factr,&pgtol,wa,iwa,task,&iprint, csave,lsave,isave,dsave); } void Program::initStructures() { f = 0; g = new double[n]; wa = new double[(2*m + 4)*n + 12*m*(m + 1)]; iwa = new int[3*n]; } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabstring.cpp000644 000765 000024 00000002007 11422240325 022461 0ustar00carlstaff000000 000000 #include "matlabstring.h" #include "matlabexception.h" // Function definitions. // ----------------------------------------------------------------- char* copystring (const char* source) { int n = strlen(source); // The length of the string. char* dest = new char[n+1]; // The return value. strcpy(dest,source); return dest; } // Function definitions for class MatlabString. // ----------------------------------------------------------------- MatlabString::MatlabString (const mxArray* ptr) { s = 0; // Check to make sure the Matlab array is a string. if (!mxIsChar(ptr)) throw MatlabException("Matlab array must be a string (of type CHAR)"); // Get the string passed as a Matlab array. s = mxArrayToString(ptr); if (s == 0) throw MatlabException("Unable to obtain string from Matlab array"); } MatlabString::MatlabString (const MatlabString& source) { s = 0; // Copy the source string. s = copystring(source.s); } MatlabString::~MatlabString() { if (s) mxFree(s); } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabexception.h000644 000765 000024 00000001545 11422240325 022624 0ustar00carlstaff000000 000000 #ifndef INCLUDE_MATLABEXCEPTION #define INCLUDE_MATLABEXCEPTION #include // Class MatlabException // ----------------------------------------------------------------- // This class just makes it easier for me to throw exceptions. Its // functionality really has nothing to do with MATLAB. class MatlabException : public std::exception { public: MatlabException (const char* message) throw(); ~MatlabException() throw() { }; // The copy constructor makes a shallow copy. MatlabException (const MatlabException& source) throw(); // The copy assignment operator makes a shallow copy as well. MatlabException& operator= (const MatlabException& source); // Return the message string. virtual const char* what () const throw() { return message; }; private: const char* message; // The error message. }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/Makefile000644 000765 000024 00000004265 11527514533 020751 0ustar00carlstaff000000 000000 # path to your Matlab installation # you can find it by the commands 'locate matlab' or 'find / -name "matlab"' MATLAB_HOME=/agbs/share/sw/matlab #MATLAB_HOME=/usr/local/share/matlabR2008b # for 'make mex' you need to provide the variable MEX and have mex set up # for 'make nomex' you need to provide the variables MEX_SUFFIX and MATLAB_LIB # Choose Operating System ###################################################### # For 32 bit Linux, uncomment the following three lines.------------------------ MEX = $(MATLAB_HOME)/bin/mex MEX_SUFFIX = mexglx MATLAB_LIB = -L$(MATLAB_HOME)/bin/glnx86 -lmex # For 64 bit Linux, uncomment the following three lines.------------------------ #MEX = $(MATLAB_HOME)/bin/mex #MEX_SUFFIX = mexa64 #MATLAB_LIB = -L$(MATLAB_HOME)/bin/glnxa64 -lmex # For Mac OS X, uncomment the following three lines.---------------------------- #MEX = /Applications/MATLAB7/bin/mex #MEX_SUFFIX = mexmac #MATLAB_LIB = -L$(MATLAB_HOME)/bin/maci -lmex # Choose FORTRAN compiler ###################################################### # To work with f77, uncomment the following two lines.-------------------------- F77 = f77 FORTRAN_LIB = g2c # To use gfortran, uncomment the following two lines.--------------------------- #F77 = gfortran #FORTRAN_LIB = gfortran # Do not edit below ############################################################ CXX = g++ CFLAGS = -O3 -ffast-math -fomit-frame-pointer -fPIC -Werror -pthread -Wall -ansi FFLAGS = -O3 -fPIC -fexceptions MATLAB_INCLUDE=-I$(MATLAB_HOME)/extern/include TARGET = lbfgsb OBJS = solver.o matlabexception.o matlabscalar.o matlabstring.o \ matlabmatrix.o arrayofmatrices.o program.o matlabprogram.o \ lbfgsb.o %.o: %.cpp $(CXX) $(CFLAGS) $(MATLAB_INCLUDE) -o $@ -c $^ %.o: %.f $(F77) $(FFLAGS) -o $@ -c $^ mex: $(TARGET) clean nomex: $(TARGET)_nomex clean $(TARGET): $(OBJS) $(MEX) -cxx CXX=$(CXX) CC=$(CXX) FC=$(FCC) LD=$(CXX) -l$(FORTRAN_LIB) -lm \ -O -output $@ $^ $(TARGET)_nomex: $(OBJS) $(CXX) $^ -shared -o $(TARGET).$(MEX_SUFFIX) $(MATLAB_LIB) -l$(FORTRAN_LIB) -lm clean: rm -f *.o cp $(TARGET).$(MEX_SUFFIX) ../$(TARGET).mex mv $(TARGET).$(MEX_SUFFIX) .. gpml-matlab-v3.1-2010-09-27/util/lbfgsb/solver.f000644 000765 000024 00000423410 11422240325 020754 0ustar00carlstaff000000 000000 c================ L-BFGS-B (version 2.4) ========================== subroutine setulb(n, m, x, l, u, nbd, f, g, factr, pgtol, wa, iwa, + task, iprint, csave, lsave, isave, dsave) character*60 task, csave logical lsave(4) integer n, m, iprint, + nbd(n), iwa(3*n), isave(44) double precision f, factr, pgtol, x(n), l(n), u(n), g(n), + wa(2*m*n+4*n+11*m*m+8*m), dsave(29) c ************ c c Subroutine setulb c c This subroutine partitions the working arrays wa and iwa, and c then uses the limited memory BFGS method to solve the bound c constrained optimization problem by calling mainlb. c (The direct method will be used in the subspace minimization.) c c n is an integer variable. c On entry n is the dimension of the problem. c On exit n is unchanged. c c m is an integer variable. c On entry m is the maximum number of variable metric corrections c used to define the limited memory matrix. c On exit m is unchanged. c c x is a double precision array of dimension n. c On entry x is an approximation to the solution. c On exit x is the current approximation. c c l is a double precision array of dimension n. c On entry l is the lower bound on x. c On exit l is unchanged. c c u is a double precision array of dimension n. c On entry u is the upper bound on x. c On exit u is unchanged. c c nbd is an integer array of dimension n. c On entry nbd represents the type of bounds imposed on the c variables, and must be specified as follows: c nbd(i)=0 if x(i) is unbounded, c 1 if x(i) has only a lower bound, c 2 if x(i) has both lower and upper bounds, and c 3 if x(i) has only an upper bound. c On exit nbd is unchanged. c c f is a double precision variable. c On first entry f is unspecified. c On final exit f is the value of the function at x. c c g is a double precision array of dimension n. c On first entry g is unspecified. c On final exit g is the value of the gradient at x. c c factr is a double precision variable. c On entry factr >= 0 is specified by the user. The iteration c will stop when c c (f^k - f^{k+1})/max{|f^k|,|f^{k+1}|,1} <= factr*epsmch c c where epsmch is the machine precision, which is automatically c generated by the code. Typical values for factr: 1.d+12 for c low accuracy; 1.d+7 for moderate accuracy; 1.d+1 for extremely c high accuracy. c On exit factr is unchanged. c c pgtol is a double precision variable. c On entry pgtol >= 0 is specified by the user. The iteration c will stop when c c max{|proj g_i | i = 1, ..., n} <= pgtol c c where pg_i is the ith component of the projected gradient. c On exit pgtol is unchanged. c c wa is a double precision working array of length c (2mmax + 4)nmax + 11mmax^2 + 8mmax. c c iwa is an integer working array of length 3nmax. c c task is a working string of characters of length 60 indicating c the current job when entering and quitting this subroutine. c c iprint is an integer variable that must be set by the user. c It controls the frequency and type of output generated: c iprint<0 no output is generated; c iprint=0 print only one line at the last iteration; c 0100 print details of every iteration including x and g; c When iprint > 0, the file iterate.dat will be created to c summarize the iteration. c c csave is a working string of characters of length 60. c c lsave is a logical working array of dimension 4. c On exit with 'task' = NEW_X, the following information is c available: c If lsave(1) = .true. then the initial X has been replaced by c its projection in the feasible set; c If lsave(2) = .true. then the problem is constrained; c If lsave(3) = .true. then each variable has upper and lower c bounds; c c isave is an integer working array of dimension 44. c On exit with 'task' = NEW_X, the following information is c available: c isave(22) = the total number of intervals explored in the c search of Cauchy points; c isave(26) = the total number of skipped BFGS updates before c the current iteration; c isave(30) = the number of current iteration; c isave(31) = the total number of BFGS updates prior the current c iteration; c isave(33) = the number of intervals explored in the search of c Cauchy point in the current iteration; c isave(34) = the total number of function and gradient c evaluations; c isave(36) = the number of function value or gradient c evaluations in the current iteration; c if isave(37) = 0 then the subspace argmin is within the box; c if isave(37) = 1 then the subspace argmin is beyond the box; c isave(38) = the number of free variables in the current c iteration; c isave(39) = the number of active constraints in the current c iteration; c n + 1 - isave(40) = the number of variables leaving the set of c active constraints in the current iteration; c isave(41) = the number of variables entering the set of active c constraints in the current iteration. c c dsave is a double precision working array of dimension 29. c On exit with 'task' = NEW_X, the following information is c available: c dsave(1) = current 'theta' in the BFGS matrix; c dsave(2) = f(x) in the previous iteration; c dsave(3) = factr*epsmch; c dsave(4) = 2-norm of the line search direction vector; c dsave(5) = the machine precision epsmch generated by the code; c dsave(7) = the accumulated time spent on searching for c Cauchy points; c dsave(8) = the accumulated time spent on c subspace minimization; c dsave(9) = the accumulated time spent on line search; c dsave(11) = the slope of the line search function at c the current point of line search; c dsave(12) = the maximum relative step length imposed in c line search; c dsave(13) = the infinity norm of the projected gradient; c dsave(14) = the relative step length in the line search; c dsave(15) = the slope of the line search function at c the starting point of the line search; c dsave(16) = the square of the 2-norm of the line search c direction vector. c c Subprograms called: c c L-BFGS-B Library ... mainlb. c c c References: c c [1] R. H. Byrd, P. Lu, J. Nocedal and C. Zhu, ``A limited c memory algorithm for bound constrained optimization'', c SIAM J. Scientific Computing 16 (1995), no. 5, pp. 1190--1208. c c [2] C. Zhu, R.H. Byrd, P. Lu, J. Nocedal, ``L-BFGS-B: a c limited memory FORTRAN code for solving bound constrained c optimization problems'', Tech. Report, NAM-11, EECS Department, c Northwestern University, 1994. c c (Postscript files of these papers are available via anonymous c ftp to ece.nwu.edu in the directory pub/lbfgs/lbfgs_bcm.) c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer l1,l2,l3,lws,lr,lz,lt,ld,lwa,lwy,lsy,lss,lwt,lwn,lsnd if (task .eq. 'START') then isave(1) = m*n isave(2) = m**2 isave(3) = 4*m**2 isave(4) = 1 isave(5) = isave(4) + isave(1) isave(6) = isave(5) + isave(1) isave(7) = isave(6) + isave(2) isave(8) = isave(7) + isave(2) isave(9) = isave(8) isave(10) = isave(9) + isave(2) isave(11) = isave(10) + isave(3) isave(12) = isave(11) + isave(3) isave(13) = isave(12) + n isave(14) = isave(13) + n isave(15) = isave(14) + n isave(16) = isave(15) + n endif l1 = isave(1) l2 = isave(2) l3 = isave(3) lws = isave(4) lwy = isave(5) lsy = isave(6) lss = isave(7) lwt = isave(9) lwn = isave(10) lsnd = isave(11) lz = isave(12) lr = isave(13) ld = isave(14) lt = isave(15) lwa = isave(16) call mainlb(n,m,x,l,u,nbd,f,g,factr,pgtol, + wa(lws),wa(lwy),wa(lsy),wa(lss),wa(lwt), + wa(lwn),wa(lsnd),wa(lz),wa(lr),wa(ld),wa(lt), + wa(lwa),iwa(1),iwa(n+1),iwa(2*n+1),task,iprint, + csave,lsave,isave(22),dsave) return end c======================= The end of setulb ============================= subroutine mainlb(n, m, x, l, u, nbd, f, g, factr, pgtol, ws, wy, + sy, ss, wt, wn, snd, z, r, d, t, wa, + index, iwhere, indx2, task, iprint, + csave, lsave, isave, dsave) character*60 task, csave logical lsave(4) integer n, m, iprint, nbd(n), index(n), + iwhere(n), indx2(n), isave(23) double precision f, factr, pgtol, + x(n), l(n), u(n), g(n), z(n), r(n), d(n), t(n), + wa(8*m), ws(n, m), wy(n, m), sy(m, m), ss(m, m), + wt(m, m), wn(2*m, 2*m), snd(2*m, 2*m), dsave(29) c ************ c c Subroutine mainlb c c This subroutine solves bound constrained optimization problems by c using the compact formula of the limited memory BFGS updates. c c n is an integer variable. c On entry n is the number of variables. c On exit n is unchanged. c c m is an integer variable. c On entry m is the maximum number of variable metric c corrections allowed in the limited memory matrix. c On exit m is unchanged. c c x is a double precision array of dimension n. c On entry x is an approximation to the solution. c On exit x is the current approximation. c c l is a double precision array of dimension n. c On entry l is the lower bound of x. c On exit l is unchanged. c c u is a double precision array of dimension n. c On entry u is the upper bound of x. c On exit u is unchanged. c c nbd is an integer array of dimension n. c On entry nbd represents the type of bounds imposed on the c variables, and must be specified as follows: c nbd(i)=0 if x(i) is unbounded, c 1 if x(i) has only a lower bound, c 2 if x(i) has both lower and upper bounds, c 3 if x(i) has only an upper bound. c On exit nbd is unchanged. c c f is a double precision variable. c On first entry f is unspecified. c On final exit f is the value of the function at x. c c g is a double precision array of dimension n. c On first entry g is unspecified. c On final exit g is the value of the gradient at x. c c factr is a double precision variable. c On entry factr >= 0 is specified by the user. The iteration c will stop when c c (f^k - f^{k+1})/max{|f^k|,|f^{k+1}|,1} <= factr*epsmch c c where epsmch is the machine precision, which is automatically c generated by the code. c On exit factr is unchanged. c c pgtol is a double precision variable. c On entry pgtol >= 0 is specified by the user. The iteration c will stop when c c max{|proj g_i | i = 1, ..., n} <= pgtol c c where pg_i is the ith component of the projected gradient. c On exit pgtol is unchanged. c c ws, wy, sy, and wt are double precision working arrays used to c store the following information defining the limited memory c BFGS matrix: c ws, of dimension n x m, stores S, the matrix of s-vectors; c wy, of dimension n x m, stores Y, the matrix of y-vectors; c sy, of dimension m x m, stores S'Y; c ss, of dimension m x m, stores S'S; c wt, of dimension m x m, stores the Cholesky factorization c of (theta*S'S+LD^(-1)L'); see eq. c (2.26) in [3]. c c wn is a double precision working array of dimension 2m x 2m c used to store the LEL^T factorization of the indefinite matrix c K = [-D -Y'ZZ'Y/theta L_a'-R_z' ] c [L_a -R_z theta*S'AA'S ] c c where E = [-I 0] c [ 0 I] c c snd is a double precision working array of dimension 2m x 2m c used to store the lower triangular part of c N = [Y' ZZ'Y L_a'+R_z'] c [L_a +R_z S'AA'S ] c c z(n),r(n),d(n),t(n),wa(8*m) are double precision working arrays. c z is used at different times to store the Cauchy point and c the Newton point. c c c index is an integer working array of dimension n. c In subroutine freev, index is used to store the free and fixed c variables at the Generalized Cauchy Point (GCP). c c iwhere is an integer working array of dimension n used to record c the status of the vector x for GCP computation. c iwhere(i)=0 or -3 if x(i) is free and has bounds, c 1 if x(i) is fixed at l(i), and l(i) .ne. u(i) c 2 if x(i) is fixed at u(i), and u(i) .ne. l(i) c 3 if x(i) is always fixed, i.e., u(i)=x(i)=l(i) c -1 if x(i) is always free, i.e., no bounds on it. c c indx2 is an integer working array of dimension n. c Within subroutine cauchy, indx2 corresponds to the array iorder. c In subroutine freev, a list of variables entering and leaving c the free set is stored in indx2, and it is passed on to c subroutine formk with this information. c c task is a working string of characters of length 60 indicating c the current job when entering and leaving this subroutine. c c iprint is an INTEGER variable that must be set by the user. c It controls the frequency and type of output generated: c iprint<0 no output is generated; c iprint=0 print only one line at the last iteration; c 0100 print details of every iteration including x and g; c When iprint > 0, the file iterate.dat will be created to c summarize the iteration. c c csave is a working string of characters of length 60. c c lsave is a logical working array of dimension 4. c c isave is an integer working array of dimension 23. c c dsave is a double precision working array of dimension 29. c c c Subprograms called c c L-BFGS-B Library ... cauchy, subsm, lnsrlb, formk, c c errclb, prn1lb, prn2lb, prn3lb, active, projgr, c c freev, cmprlb, matupd, formt. c c Minpack2 Library ... timer, dpmeps. c c Linpack Library ... dcopy, ddot. c c c References: c c [1] R. H. Byrd, P. Lu, J. Nocedal and C. Zhu, ``A limited c memory algorithm for bound constrained optimization'', c SIAM J. Scientific Computing 16 (1995), no. 5, pp. 1190--1208. c c [2] C. Zhu, R.H. Byrd, P. Lu, J. Nocedal, ``L-BFGS-B: FORTRAN c Subroutines for Large Scale Bound Constrained Optimization'' c Tech. Report, NAM-11, EECS Department, Northwestern University, c 1994. c c [3] R. Byrd, J. Nocedal and R. Schnabel "Representations of c Quasi-Newton Matrices and their use in Limited Memory Methods'', c Mathematical Programming 63 (1994), no. 4, pp. 129-156. c c (Postscript files of these papers are available via anonymous c ftp to ece.nwu.edu in the directory pub/lbfgs/lbfgs_bcm.) c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ logical prjctd,cnstnd,boxed,updatd,wrk character*3 word integer i,k,nintol,itfile,iback,nskip, + head,col,iter,itail,iupdat, + nint,nfgv,info,ifun, + iword,nfree,nact,ileave,nenter double precision theta,fold,ddot,dr,rr,tol,dpmeps, + xstep,sbgnrm,ddum,dnorm,dtd,epsmch, + cpu1,cpu2,cachyt,sbtime,lnscht,time1,time2, + gd,gdold,stp,stpmx,time double precision one,zero parameter (one=1.0d0,zero=0.0d0) if (task .eq. 'START') then call timer(time1) c Generate the current machine precision. epsmch = dpmeps() fold = 0.0d0 dnorm = 0.0d0 cpu1 = 0.0d0 gd = 0.0d0 sbgnrm = 0.0d0 stp = 0.0d0 stpmx = 0.0d0 gdold = 0.0d0 dtd = 0.0d0 c Initialize counters and scalars when task='START'. c for the limited memory BFGS matrices: col = 0 head = 1 theta = one iupdat = 0 updatd = .false. iback = 0 itail = 0 ifun = 0 iword = 0 nact = 0 ileave = 0 nenter = 0 c for operation counts: iter = 0 nfgv = 0 nint = 0 nintol = 0 nskip = 0 nfree = n c for stopping tolerance: tol = factr*epsmch c for measuring running time: cachyt = 0 sbtime = 0 lnscht = 0 c 'word' records the status of subspace solutions. word = '---' c 'info' records the termination information. info = 0 itfile = 0 if (iprint .ge. 1) then c open a summary file 'iterate.dat' open (8, file = 'iterate.dat', status = 'unknown') itfile = 8 endif c Check the input arguments for errors. call errclb(n,m,factr,l,u,nbd,task,info,k) if (task(1:5) .eq. 'ERROR') then call prn3lb(n,x,f,task,iprint,info,itfile, + iter,nfgv,nintol,nskip,nact,sbgnrm, + zero,nint,word,iback,stp,xstep,k, + cachyt,sbtime,lnscht) return endif call prn1lb(n,m,l,u,x,iprint,itfile,epsmch) c Initialize iwhere & project x onto the feasible set. call active(n,l,u,nbd,x,iwhere,iprint,prjctd,cnstnd,boxed) c The end of the initialization. else c restore local variables. prjctd = lsave(1) cnstnd = lsave(2) boxed = lsave(3) updatd = lsave(4) nintol = isave(1) itfile = isave(3) iback = isave(4) nskip = isave(5) head = isave(6) col = isave(7) itail = isave(8) iter = isave(9) iupdat = isave(10) nint = isave(12) nfgv = isave(13) info = isave(14) ifun = isave(15) iword = isave(16) nfree = isave(17) nact = isave(18) ileave = isave(19) nenter = isave(20) theta = dsave(1) fold = dsave(2) tol = dsave(3) dnorm = dsave(4) epsmch = dsave(5) cpu1 = dsave(6) cachyt = dsave(7) sbtime = dsave(8) lnscht = dsave(9) time1 = dsave(10) gd = dsave(11) stpmx = dsave(12) sbgnrm = dsave(13) stp = dsave(14) gdold = dsave(15) dtd = dsave(16) c After returning from the driver go to the point where execution c is to resume. if (task(1:5) .eq. 'FG_LN') goto 666 if (task(1:5) .eq. 'NEW_X') goto 777 if (task(1:5) .eq. 'FG_ST') goto 111 if (task(1:4) .eq. 'STOP') then if (task(7:9) .eq. 'CPU') then c restore the previous iterate. call dcopy(n,t,1,x,1) call dcopy(n,r,1,g,1) f = fold endif goto 999 endif endif c Compute f0 and g0. task = 'FG_START' c return to the driver to calculate f and g; reenter at 111. goto 1000 111 continue nfgv = 1 c Compute the infinity norm of the (-) projected gradient. call projgr(n,l,u,nbd,x,g,sbgnrm) if (iprint .ge. 1) then write (6,1002) iter,f,sbgnrm write (itfile,1003) iter,nfgv,sbgnrm,f endif if (sbgnrm .le. pgtol) then c terminate the algorithm. task = 'CONVERGENCE: NORM OF PROJECTED GRADIENT <= PGTOL' goto 999 endif c ----------------- the beginning of the loop -------------------------- 222 continue if (iprint .ge. 99) write (6,1001) iter + 1 iword = -1 c if (.not. cnstnd .and. col .gt. 0) then c skip the search for GCP. call dcopy(n,x,1,z,1) wrk = updatd nint = 0 goto 333 endif cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc c c Compute the Generalized Cauchy Point (GCP). c cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc call timer(cpu1) call cauchy(n,x,l,u,nbd,g,indx2,iwhere,t,d,z, + m,wy,ws,sy,wt,theta,col,head, + wa(1),wa(2*m+1),wa(4*m+1),wa(6*m+1),nint, + iprint,sbgnrm,info,epsmch) if (info .ne. 0) then c singular triangular system detected; refresh the lbfgs memory. if(iprint .ge. 1) write (6, 1005) info = 0 col = 0 head = 1 theta = one iupdat = 0 updatd = .false. call timer(cpu2) cachyt = cachyt + cpu2 - cpu1 goto 222 endif call timer(cpu2) cachyt = cachyt + cpu2 - cpu1 nintol = nintol + nint c Count the entering and leaving variables for iter > 0; c find the index set of free and active variables at the GCP. call freev(n,nfree,index,nenter,ileave,indx2, + iwhere,wrk,updatd,cnstnd,iprint,iter) nact = n - nfree 333 continue c If there are no free variables or B=theta*I, then c skip the subspace minimization. if (nfree .eq. 0 .or. col .eq. 0) goto 555 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc c c Subspace minimization. c cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc call timer(cpu1) c Form the LEL^T factorization of the indefinite c matrix K = [-D -Y'ZZ'Y/theta L_a'-R_z' ] c [L_a -R_z theta*S'AA'S ] c where E = [-I 0] c [ 0 I] if (wrk) call formk(n,nfree,index,nenter,ileave,indx2,iupdat, + updatd,wn,snd,m,ws,wy,sy,theta,col,head,info) if (info .ne. 0) then c nonpositive definiteness in Cholesky factorization; c refresh the lbfgs memory and restart the iteration. if(iprint .ge. 1) write (6, 1006) info = 0 col = 0 head = 1 theta = one iupdat = 0 updatd = .false. call timer(cpu2) sbtime = sbtime + cpu2 - cpu1 goto 222 endif c compute r=-Z'B(xcp-xk)-Z'g (using wa(2m+1)=W'(xcp-x) c from 'cauchy'). call cmprlb(n,m,x,g,ws,wy,sy,wt,z,r,wa,index, + theta,col,head,nfree,cnstnd,info) if (info .ne. 0) goto 444 c call the direct method. call subsm(n,m,nfree,index,l,u,nbd,z,r,ws,wy,theta, + col,head,iword,wa,wn,iprint,info) 444 continue if (info .ne. 0) then c singular triangular system detected; c refresh the lbfgs memory and restart the iteration. if(iprint .ge. 1) write (6, 1005) info = 0 col = 0 head = 1 theta = one iupdat = 0 updatd = .false. call timer(cpu2) sbtime = sbtime + cpu2 - cpu1 goto 222 endif call timer(cpu2) sbtime = sbtime + cpu2 - cpu1 555 continue cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc c c Line search and optimality tests. c cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc c Generate the search direction d:=z-x. do 40 i = 1, n d(i) = z(i) - x(i) 40 continue call timer(cpu1) 666 continue call lnsrlb(n,l,u,nbd,x,f,fold,gd,gdold,g,d,r,t,z,stp,dnorm, + dtd,xstep,stpmx,iter,ifun,iback,nfgv,info,task, + boxed,cnstnd,csave,isave(22),dsave(17)) if (info .ne. 0 .or. iback .ge. 20) then c restore the previous iterate. call dcopy(n,t,1,x,1) call dcopy(n,r,1,g,1) f = fold if (col .eq. 0) then c abnormal termination. if (info .eq. 0) then info = -9 c restore the actual number of f and g evaluations etc. nfgv = nfgv - 1 ifun = ifun - 1 iback = iback - 1 endif task = 'ABNORMAL_TERMINATION_IN_LNSRCH' iter = iter + 1 goto 999 else c refresh the lbfgs memory and restart the iteration. if(iprint .ge. 1) write (6, 1008) if (info .eq. 0) nfgv = nfgv - 1 info = 0 col = 0 head = 1 theta = one iupdat = 0 updatd = .false. task = 'RESTART_FROM_LNSRCH' call timer(cpu2) lnscht = lnscht + cpu2 - cpu1 goto 222 endif else if (task(1:5) .eq. 'FG_LN') then c return to the driver for calculating f and g; reenter at 666. goto 1000 else c calculate and print out the quantities related to the new X. call timer(cpu2) lnscht = lnscht + cpu2 - cpu1 iter = iter + 1 c Compute the infinity norm of the projected (-)gradient. call projgr(n,l,u,nbd,x,g,sbgnrm) c Print iteration information. call prn2lb(n,x,f,g,iprint,itfile,iter,nfgv,nact, + sbgnrm,nint,word,iword,iback,stp,xstep) goto 1000 endif 777 continue c Test for termination. if (sbgnrm .le. pgtol) then c terminate the algorithm. task = 'CONVERGENCE: NORM OF PROJECTED GRADIENT <= PGTOL' goto 999 endif ddum = max(abs(fold), abs(f), one) if ((fold - f) .le. tol*ddum) then c terminate the algorithm. task = 'CONVERGENCE: REL_REDUCTION_OF_F <= FACTR*EPSMCH' if (iback .ge. 10) info = -5 c i.e., to issue a warning if iback>10 in the line search. goto 999 endif c Compute d=newx-oldx, r=newg-oldg, rr=y'y and dr=y's. do 42 i = 1, n r(i) = g(i) - r(i) 42 continue rr = ddot(n,r,1,r,1) if (stp .eq. one) then dr = gd - gdold ddum = -gdold else dr = (gd - gdold)*stp call dscal(n,stp,d,1) ddum = -gdold*stp endif if (dr .le. epsmch*ddum) then c skip the L-BFGS update. nskip = nskip + 1 updatd = .false. if (iprint .ge. 1) write (6,1004) dr, ddum goto 888 endif cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc c c Update the L-BFGS matrix. c cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc updatd = .true. iupdat = iupdat + 1 c Update matrices WS and WY and form the middle matrix in B. call matupd(n,m,ws,wy,sy,ss,d,r,itail, + iupdat,col,head,theta,rr,dr,stp,dtd) c Form the upper half of the pds T = theta*SS + L*D^(-1)*L'; c Store T in the upper triangular of the array wt; c Cholesky factorize T to J*J' with c J' stored in the upper triangular of wt. call formt(m,wt,sy,ss,col,theta,info) if (info .ne. 0) then c nonpositive definiteness in Cholesky factorization; c refresh the lbfgs memory and restart the iteration. if(iprint .ge. 1) write (6, 1007) info = 0 col = 0 head = 1 theta = one iupdat = 0 updatd = .false. goto 222 endif c Now the inverse of the middle matrix in B is c [ D^(1/2) O ] [ -D^(1/2) D^(-1/2)*L' ] c [ -L*D^(-1/2) J ] [ 0 J' ] 888 continue c -------------------- the end of the loop ----------------------------- goto 222 999 continue call timer(time2) time = time2 - time1 call prn3lb(n,x,f,task,iprint,info,itfile, + iter,nfgv,nintol,nskip,nact,sbgnrm, + time,nint,word,iback,stp,xstep,k, + cachyt,sbtime,lnscht) 1000 continue c Save local variables. lsave(1) = prjctd lsave(2) = cnstnd lsave(3) = boxed lsave(4) = updatd isave(1) = nintol isave(3) = itfile isave(4) = iback isave(5) = nskip isave(6) = head isave(7) = col isave(8) = itail isave(9) = iter isave(10) = iupdat isave(12) = nint isave(13) = nfgv isave(14) = info isave(15) = ifun isave(16) = iword isave(17) = nfree isave(18) = nact isave(19) = ileave isave(20) = nenter dsave(1) = theta dsave(2) = fold dsave(3) = tol dsave(4) = dnorm dsave(5) = epsmch dsave(6) = cpu1 dsave(7) = cachyt dsave(8) = sbtime dsave(9) = lnscht dsave(10) = time1 dsave(11) = gd dsave(12) = stpmx dsave(13) = sbgnrm dsave(14) = stp dsave(15) = gdold dsave(16) = dtd 1001 format (//,'ITERATION ',i5) 1002 format + (/,'At iterate',i5,4x,'f= ',1p,d12.5,4x,'|proj g|= ',1p,d12.5) 1003 format (2(1x,i4),5x,'-',5x,'-',3x,'-',5x,'-',5x,'-',8x,'-',3x, + 1p,2(1x,d10.3)) 1004 format (' ys=',1p,e10.3,' -gs=',1p,e10.3,' BFGS update SKIPPED') 1005 format (/, +' Singular triangular system detected;',/, +' refresh the lbfgs memory and restart the iteration.') 1006 format (/, +' Nonpositive definiteness in Cholesky factorization in formk;',/, +' refresh the lbfgs memory and restart the iteration.') 1007 format (/, +' Nonpositive definiteness in Cholesky factorization in formt;',/, +' refresh the lbfgs memory and restart the iteration.') 1008 format (/, +' Bad direction in the line search;',/, +' refresh the lbfgs memory and restart the iteration.') return end c======================= The end of mainlb ============================= subroutine active(n, l, u, nbd, x, iwhere, iprint, + prjctd, cnstnd, boxed) logical prjctd, cnstnd, boxed integer n, iprint, nbd(n), iwhere(n) double precision x(n), l(n), u(n) c ************ c c Subroutine active c c This subroutine initializes iwhere and projects the initial x to c the feasible set if necessary. c c iwhere is an integer array of dimension n. c On entry iwhere is unspecified. c On exit iwhere(i)=-1 if x(i) has no bounds c 3 if l(i)=u(i) c 0 otherwise. c In cauchy, iwhere is given finer gradations. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer nbdd,i double precision zero parameter (zero=0.0d0) c Initialize nbdd, prjctd, cnstnd and boxed. nbdd = 0 prjctd = .false. cnstnd = .false. boxed = .true. c Project the initial x to the easible set if necessary. do 10 i = 1, n if (nbd(i) .gt. 0) then if (nbd(i) .le. 2 .and. x(i) .le. l(i)) then if (x(i) .lt. l(i)) then prjctd = .true. x(i) = l(i) endif nbdd = nbdd + 1 else if (nbd(i) .ge. 2 .and. x(i) .ge. u(i)) then if (x(i) .gt. u(i)) then prjctd = .true. x(i) = u(i) endif nbdd = nbdd + 1 endif endif 10 continue c Initialize iwhere and assign values to cnstnd and boxed. do 20 i = 1, n if (nbd(i) .ne. 2) boxed = .false. if (nbd(i) .eq. 0) then c this variable is always free iwhere(i) = -1 c otherwise set x(i)=mid(x(i), u(i), l(i)). else cnstnd = .true. if (nbd(i) .eq. 2 .and. u(i) - l(i) .le. zero) then c this variable is always fixed iwhere(i) = 3 else iwhere(i) = 0 endif endif 20 continue if (iprint .ge. 0) then if (prjctd) write (6,*) + 'The initial X is infeasible. Restart with its projection.' if (.not. cnstnd) + write (6,*) 'This problem is unconstrained.' endif if (iprint .gt. 0) write (6,1001) nbdd 1001 format (/,'At X0 ',i9,' variables are exactly at the bounds') return end c======================= The end of active ============================= subroutine bmv(m, sy, wt, col, v, p, info) integer m, col, info double precision sy(m, m), wt(m, m), v(2*col), p(2*col) c ************ c c Subroutine bmv c c This subroutine computes the product of the 2m x 2m middle matrix c in the compact L-BFGS formula of B and a 2m vector v; c it returns the product in p. c c m is an integer variable. c On entry m is the maximum number of variable metric corrections c used to define the limited memory matrix. c On exit m is unchanged. c c sy is a double precision array of dimension m x m. c On entry sy specifies the matrix S'Y. c On exit sy is unchanged. c c wt is a double precision array of dimension m x m. c On entry wt specifies the upper triangular matrix J' which is c the Cholesky factor of (thetaS'S+LD^(-1)L'). c On exit wt is unchanged. c c col is an integer variable. c On entry col specifies the number of s-vectors (or y-vectors) c stored in the compact L-BFGS formula. c On exit col is unchanged. c c v is a double precision array of dimension 2col. c On entry v specifies vector v. c On exit v is unchanged. c c p is a double precision array of dimension 2col. c On entry p is unspecified. c On exit p is the product Mv. c c info is an integer variable. c On entry info is unspecified. c On exit info = 0 for normal return, c = nonzero for abnormal return when the system c to be solved by dtrsl is singular. c c Subprograms called: c c Linpack ... dtrsl. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i,k,i2 double precision sum if (col .eq. 0) return c PART I: solve [ D^(1/2) O ] [ p1 ] = [ v1 ] c [ -L*D^(-1/2) J ] [ p2 ] [ v2 ]. c solve Jp2=v2+LD^(-1)v1. p(col + 1) = v(col + 1) do 20 i = 2, col i2 = col + i sum = 0.0d0 do 10 k = 1, i - 1 sum = sum + sy(i,k)*v(k)/sy(k,k) 10 continue p(i2) = v(i2) + sum 20 continue c Solve the triangular system call dtrsl(wt,m,col,p(col+1),11,info) if (info .ne. 0) return c solve D^(1/2)p1=v1. cc do 30 i = 1, col cc p(i) = v(i)/sqrt(sy(i,i)) cc30 continue c PART II: solve [ -D^(1/2) D^(-1/2)*L' ] [ p1 ] = [ p1 ] c [ 0 J' ] [ p2 ] [ p2 ]. c solve J^Tp2=p2. call dtrsl(wt,m,col,p(col+1),01,info) if (info .ne. 0) return c compute p1=-D^(-1/2)(p1-D^(-1/2)L'p2) c =-D^(-1/2)p1+D^(-1)L'p2. do 40 i = 1, col cc p(i) = -p(i)/sqrt(sy(i,i)) combined with do 30 loop cc into the next line p(i) = -v(i)/sy(i,i) 40 continue do 60 i = 1, col sum = 0.d0 do 50 k = i + 1, col sum = sum + sy(k,i)*p(col+k)/sy(i,i) 50 continue p(i) = p(i) + sum 60 continue return end c======================== The end of bmv =============================== subroutine cauchy(n, x, l, u, nbd, g, iorder, iwhere, t, d, xcp, + m, wy, ws, sy, wt, theta, col, head, p, c, wbp, + v, nint, iprint, sbgnrm, info, epsmch) integer n, m, head, col, nint, iprint, info, + nbd(n), iorder(n), iwhere(n) double precision theta, epsmch, + x(n), l(n), u(n), g(n), t(n), d(n), xcp(n), + wy(n, col), ws(n, col), sy(m, m), + wt(m, m), p(2*m), c(2*m), wbp(2*m), v(2*m) c ************ c c Subroutine cauchy c c For given x, l, u, g (with sbgnrm > 0), and a limited memory c BFGS matrix B defined in terms of matrices WY, WS, WT, and c scalars head, col, and theta, this subroutine computes the c generalized Cauchy point (GCP), defined as the first local c minimizer of the quadratic c c Q(x + s) = g's + 1/2 s'Bs c c along the projected gradient direction P(x-tg,l,u). c The routine returns the GCP in xcp. c c n is an integer variable. c On entry n is the dimension of the problem. c On exit n is unchanged. c c x is a double precision array of dimension n. c On entry x is the starting point for the GCP computation. c On exit x is unchanged. c c l is a double precision array of dimension n. c On entry l is the lower bound of x. c On exit l is unchanged. c c u is a double precision array of dimension n. c On entry u is the upper bound of x. c On exit u is unchanged. c c nbd is an integer array of dimension n. c On entry nbd represents the type of bounds imposed on the c variables, and must be specified as follows: c nbd(i)=0 if x(i) is unbounded, c 1 if x(i) has only a lower bound, c 2 if x(i) has both lower and upper bounds, and c 3 if x(i) has only an upper bound. c On exit nbd is unchanged. c c g is a double precision array of dimension n. c On entry g is the gradient of f(x). g must be a nonzero vector. c On exit g is unchanged. c c iorder is an integer working array of dimension n. c iorder will be used to store the breakpoints in the piecewise c linear path and free variables encountered. On exit, c iorder(1),...,iorder(nleft) are indices of breakpoints c which have not been encountered; c iorder(nleft+1),...,iorder(nbreak) are indices of c encountered breakpoints; and c iorder(nfree),...,iorder(n) are indices of variables which c have no bound constraits along the search direction. c c iwhere is an integer array of dimension n. c On entry iwhere indicates only the permanently fixed (iwhere=3) c or free (iwhere= -1) components of x. c On exit iwhere records the status of the current x variables. c iwhere(i)=-3 if x(i) is free and has bounds, but is not moved c 0 if x(i) is free and has bounds, and is moved c 1 if x(i) is fixed at l(i), and l(i) .ne. u(i) c 2 if x(i) is fixed at u(i), and u(i) .ne. l(i) c 3 if x(i) is always fixed, i.e., u(i)=x(i)=l(i) c -1 if x(i) is always free, i.e., it has no bounds. c c t is a double precision working array of dimension n. c t will be used to store the break points. c c d is a double precision array of dimension n used to store c the Cauchy direction P(x-tg)-x. c c xcp is a double precision array of dimension n used to return the c GCP on exit. c c m is an integer variable. c On entry m is the maximum number of variable metric corrections c used to define the limited memory matrix. c On exit m is unchanged. c c ws, wy, sy, and wt are double precision arrays. c On entry they store information that defines the c limited memory BFGS matrix: c ws(n,m) stores S, a set of s-vectors; c wy(n,m) stores Y, a set of y-vectors; c sy(m,m) stores S'Y; c wt(m,m) stores the c Cholesky factorization of (theta*S'S+LD^(-1)L'). c On exit these arrays are unchanged. c c theta is a double precision variable. c On entry theta is the scaling factor specifying B_0 = theta I. c On exit theta is unchanged. c c col is an integer variable. c On entry col is the actual number of variable metric c corrections stored so far. c On exit col is unchanged. c c head is an integer variable. c On entry head is the location of the first s-vector c (or y-vector) in S (or Y). c On exit col is unchanged. c c p is a double precision working array of dimension 2m. c p will be used to store the vector p = W^(T)d. c c c is a double precision working array of dimension 2m. c c will be used to store the vector c = W^(T)(xcp-x). c c wbp is a double precision working array of dimension 2m. c wbp will be used to store the row of W corresponding c to a breakpoint. c c v is a double precision working array of dimension 2m. c c nint is an integer variable. c On exit nint records the number of quadratic segments explored c in searching for the GCP. c c iprint is an INTEGER variable that must be set by the user. c It controls the frequency and type of output generated: c iprint<0 no output is generated; c iprint=0 print only one line at the last iteration; c 0100 print details of every iteration including x and g; c When iprint > 0, the file iterate.dat will be created to c summarize the iteration. c c sbgnrm is a double precision variable. c On entry sbgnrm is the norm of the projected gradient at x. c On exit sbgnrm is unchanged. c c info is an integer variable. c On entry info is 0. c On exit info = 0 for normal return, c = nonzero for abnormal return when the the system c used in routine bmv is singular. c c Subprograms called: c c L-BFGS-B Library ... hpsolb, bmv. c c Linpack ... dscal dcopy, daxpy. c c c References: c c [1] R. H. Byrd, P. Lu, J. Nocedal and C. Zhu, ``A limited c memory algorithm for bound constrained optimization'', c SIAM J. Scientific Computing 16 (1995), no. 5, pp. 1190--1208. c c [2] C. Zhu, R.H. Byrd, P. Lu, J. Nocedal, ``L-BFGS-B: FORTRAN c Subroutines for Large Scale Bound Constrained Optimization'' c Tech. Report, NAM-11, EECS Department, Northwestern University, c 1994. c c (Postscript files of these papers are available via anonymous c ftp to ece.nwu.edu in the directory pub/lbfgs/lbfgs_bcm.) c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ logical xlower,xupper,bnded integer i,j,col2,nfree,nbreak,pointr, + ibp,nleft,ibkmin,iter double precision f1,f2,dt,dtm,tsum,dibp,zibp,dibp2,bkmin, + tu,tl,wmc,wmp,wmw,ddot,tj,tj0,neggi,sbgnrm, + f2_org double precision one,zero parameter (one=1.0d0,zero=0.0d0) c Check the status of the variables, reset iwhere(i) if necessary; c compute the Cauchy direction d and the breakpoints t; initialize c the derivative f1 and the vector p = W'd (for theta = 1). if (sbgnrm .le. zero) then if (iprint .ge. 0) write (6,*) 'Subgnorm = 0. GCP = X.' call dcopy(n,x,1,xcp,1) return endif bnded = .true. nfree = n + 1 nbreak = 0 ibkmin = 0 bkmin = zero col2 = 2*col f1 = zero if (iprint .ge. 99) write (6,3010) c We set p to zero and build it up as we determine d. do 20 i = 1, col2 p(i) = zero 20 continue c In the following loop we determine for each variable its bound c status and its breakpoint, and update p accordingly. c Smallest breakpoint is identified. do 50 i = 1, n neggi = -g(i) if (iwhere(i) .ne. 3 .and. iwhere(i) .ne. -1) then c if x(i) is not a constant and has bounds, c compute the difference between x(i) and its bounds. if (nbd(i) .le. 2) tl = x(i) - l(i) if (nbd(i) .ge. 2) tu = u(i) - x(i) c If a variable is close enough to a bound c we treat it as at bound. xlower = nbd(i) .le. 2 .and. tl .le. zero xupper = nbd(i) .ge. 2 .and. tu .le. zero c reset iwhere(i). iwhere(i) = 0 if (xlower) then if (neggi .le. zero) iwhere(i) = 1 else if (xupper) then if (neggi .ge. zero) iwhere(i) = 2 else if (abs(neggi) .le. zero) iwhere(i) = -3 endif endif pointr = head if (iwhere(i) .ne. 0 .and. iwhere(i) .ne. -1) then d(i) = zero else d(i) = neggi f1 = f1 - neggi*neggi c calculate p := p - W'e_i* (g_i). do 40 j = 1, col p(j) = p(j) + wy(i,pointr)* neggi p(col + j) = p(col + j) + ws(i,pointr)*neggi pointr = mod(pointr,m) + 1 40 continue if (nbd(i) .le. 2 .and. nbd(i) .ne. 0 + .and. neggi .lt. zero) then c x(i) + d(i) is bounded; compute t(i). nbreak = nbreak + 1 iorder(nbreak) = i t(nbreak) = tl/(-neggi) if (nbreak .eq. 1 .or. t(nbreak) .lt. bkmin) then bkmin = t(nbreak) ibkmin = nbreak endif else if (nbd(i) .ge. 2 .and. neggi .gt. zero) then c x(i) + d(i) is bounded; compute t(i). nbreak = nbreak + 1 iorder(nbreak) = i t(nbreak) = tu/neggi if (nbreak .eq. 1 .or. t(nbreak) .lt. bkmin) then bkmin = t(nbreak) ibkmin = nbreak endif else c x(i) + d(i) is not bounded. nfree = nfree - 1 iorder(nfree) = i if (abs(neggi) .gt. zero) bnded = .false. endif endif 50 continue c The indices of the nonzero components of d are now stored c in iorder(1),...,iorder(nbreak) and iorder(nfree),...,iorder(n). c The smallest of the nbreak breakpoints is in t(ibkmin)=bkmin. if (theta .ne. one) then c complete the initialization of p for theta not= one. call dscal(col,theta,p(col+1),1) endif c Initialize GCP xcp = x. call dcopy(n,x,1,xcp,1) if (nbreak .eq. 0 .and. nfree .eq. n + 1) then c is a zero vector, return with the initial xcp as GCP. if (iprint .gt. 100) write (6,1010) (xcp(i), i = 1, n) return endif c Initialize c = W'(xcp - x) = 0. do 60 j = 1, col2 c(j) = zero 60 continue c Initialize derivative f2. f2 = -theta*f1 f2_org = f2 if (col .gt. 0) then call bmv(m,sy,wt,col,p,v,info) if (info .ne. 0) return f2 = f2 - ddot(col2,v,1,p,1) endif dtm = -f1/f2 tsum = zero nint = 1 if (iprint .ge. 99) + write (6,*) 'There are ',nbreak,' breakpoints ' c If there are no breakpoints, locate the GCP and return. if (nbreak .eq. 0) goto 888 nleft = nbreak iter = 1 tj = zero c------------------- the beginning of the loop ------------------------- 777 continue c Find the next smallest breakpoint; c compute dt = t(nleft) - t(nleft + 1). tj0 = tj if (iter .eq. 1) then c Since we already have the smallest breakpoint we need not do c heapsort yet. Often only one breakpoint is used and the c cost of heapsort is avoided. tj = bkmin ibp = iorder(ibkmin) else if (iter .eq. 2) then c Replace the already used smallest breakpoint with the c breakpoint numbered nbreak > nlast, before heapsort call. if (ibkmin .ne. nbreak) then t(ibkmin) = t(nbreak) iorder(ibkmin) = iorder(nbreak) endif c Update heap structure of breakpoints c (if iter=2, initialize heap). endif call hpsolb(nleft,t,iorder,iter-2) tj = t(nleft) ibp = iorder(nleft) endif dt = tj - tj0 if (dt .ne. zero .and. iprint .ge. 100) then write (6,4011) nint,f1,f2 write (6,5010) dt write (6,6010) dtm endif c If a minimizer is within this interval, c locate the GCP and return. if (dtm .lt. dt) goto 888 c Otherwise fix one variable and c reset the corresponding component of d to zero. tsum = tsum + dt nleft = nleft - 1 iter = iter + 1 dibp = d(ibp) d(ibp) = zero if (dibp .gt. zero) then zibp = u(ibp) - x(ibp) xcp(ibp) = u(ibp) iwhere(ibp) = 2 else zibp = l(ibp) - x(ibp) xcp(ibp) = l(ibp) iwhere(ibp) = 1 endif if (iprint .ge. 100) write (6,*) 'Variable ',ibp,' is fixed.' if (nleft .eq. 0 .and. nbreak .eq. n) then c all n variables are fixed, c return with xcp as GCP. dtm = dt goto 999 endif c Update the derivative information. nint = nint + 1 dibp2 = dibp**2 c Update f1 and f2. c temporarily set f1 and f2 for col=0. f1 = f1 + dt*f2 + dibp2 - theta*dibp*zibp f2 = f2 - theta*dibp2 if (col .gt. 0) then c update c = c + dt*p. call daxpy(col2,dt,p,1,c,1) c choose wbp, c the row of W corresponding to the breakpoint encountered. pointr = head do 70 j = 1,col wbp(j) = wy(ibp,pointr) wbp(col + j) = theta*ws(ibp,pointr) pointr = mod(pointr,m) + 1 70 continue c compute (wbp)Mc, (wbp)Mp, and (wbp)M(wbp)'. call bmv(m,sy,wt,col,wbp,v,info) if (info .ne. 0) return wmc = ddot(col2,c,1,v,1) wmp = ddot(col2,p,1,v,1) wmw = ddot(col2,wbp,1,v,1) c update p = p - dibp*wbp. call daxpy(col2,-dibp,wbp,1,p,1) c complete updating f1 and f2 while col > 0. f1 = f1 + dibp*wmc f2 = f2 + 2.0d0*dibp*wmp - dibp2*wmw endif f2 = max(epsmch*f2_org,f2) if (nleft .gt. 0) then dtm = -f1/f2 goto 777 c to repeat the loop for unsearched intervals. else if(bnded) then f1 = zero f2 = zero dtm = zero else dtm = -f1/f2 endif c------------------- the end of the loop ------------------------------- 888 continue if (iprint .ge. 99) then write (6,*) write (6,*) 'GCP found in this segment' write (6,4010) nint,f1,f2 write (6,6010) dtm endif if (dtm .le. zero) dtm = zero tsum = tsum + dtm c Move free variables (i.e., the ones w/o breakpoints) and c the variables whose breakpoints haven't been reached. call daxpy(n,tsum,d,1,xcp,1) 999 continue c Update c = c + dtm*p = W'(x^c - x) c which will be used in computing r = Z'(B(x^c - x) + g). if (col .gt. 0) call daxpy(col2,dtm,p,1,c,1) if (iprint .gt. 100) write (6,1010) (xcp(i),i = 1,n) if (iprint .ge. 99) write (6,2010) 1010 format ('Cauchy X = ',/,(4x,1p,6(1x,d11.4))) 2010 format (/,'---------------- exit CAUCHY----------------------',/) 3010 format (/,'---------------- CAUCHY entered-------------------') 4010 format ('Piece ',i3,' --f1, f2 at start point ',1p,2(1x,d11.4)) 4011 format (/,'Piece ',i3,' --f1, f2 at start point ', + 1p,2(1x,d11.4)) 5010 format ('Distance to the next break point = ',1p,d11.4) 6010 format ('Distance to the stationary point = ',1p,d11.4) return end c====================== The end of cauchy ============================== subroutine cmprlb(n, m, x, g, ws, wy, sy, wt, z, r, wa, index, + theta, col, head, nfree, cnstnd, info) logical cnstnd integer n, m, col, head, nfree, info, index(n) double precision theta, + x(n), g(n), z(n), r(n), wa(4*m), + ws(n, m), wy(n, m), sy(m, m), wt(m, m) c ************ c c Subroutine cmprlb c c This subroutine computes r=-Z'B(xcp-xk)-Z'g by using c wa(2m+1)=W'(xcp-x) from subroutine cauchy. c c Subprograms called: c c L-BFGS-B Library ... bmv. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i,j,k,pointr double precision a1,a2 if (.not. cnstnd .and. col .gt. 0) then do 26 i = 1, n r(i) = -g(i) 26 continue else do 30 i = 1, nfree k = index(i) r(i) = -theta*(z(k) - x(k)) - g(k) 30 continue call bmv(m,sy,wt,col,wa(2*m+1),wa(1),info) if (info .ne. 0) then info = -8 return endif pointr = head do 34 j = 1, col a1 = wa(j) a2 = theta*wa(col + j) do 32 i = 1, nfree k = index(i) r(i) = r(i) + wy(k,pointr)*a1 + ws(k,pointr)*a2 32 continue pointr = mod(pointr,m) + 1 34 continue endif return end c======================= The end of cmprlb ============================= subroutine errclb(n, m, factr, l, u, nbd, task, info, k) character*60 task integer n, m, info, k, nbd(n) double precision factr, l(n), u(n) c ************ c c Subroutine errclb c c This subroutine checks the validity of the input data. c c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i double precision zero parameter (zero=0.0d0) c Check the input arguments for errors. if (n .le. 0) task = 'ERROR: N .LE. 0' if (m .le. 0) task = 'ERROR: M .LE. 0' if (factr .lt. zero) task = 'ERROR: FACTR .LT. 0' c Check the validity of the arrays nbd(i), u(i), and l(i). do 10 i = 1, n if (nbd(i) .lt. 0 .or. nbd(i) .gt. 3) then c return task = 'ERROR: INVALID NBD' info = -6 k = i endif if (nbd(i) .eq. 2) then if (l(i) .gt. u(i)) then c return task = 'ERROR: NO FEASIBLE SOLUTION' info = -7 k = i endif endif 10 continue return end c======================= The end of errclb ============================= subroutine formk(n, nsub, ind, nenter, ileave, indx2, iupdat, + updatd, wn, wn1, m, ws, wy, sy, theta, col, + head, info) integer n, nsub, m, col, head, nenter, ileave, iupdat, + info, ind(n), indx2(n) double precision theta, wn(2*m, 2*m), wn1(2*m, 2*m), + ws(n, m), wy(n, m), sy(m, m) logical updatd c ************ c c Subroutine formk c c This subroutine forms the LEL^T factorization of the indefinite c c matrix K = [-D -Y'ZZ'Y/theta L_a'-R_z' ] c [L_a -R_z theta*S'AA'S ] c where E = [-I 0] c [ 0 I] c The matrix K can be shown to be equal to the matrix M^[-1]N c occurring in section 5.1 of [1], as well as to the matrix c Mbar^[-1] Nbar in section 5.3. c c n is an integer variable. c On entry n is the dimension of the problem. c On exit n is unchanged. c c nsub is an integer variable c On entry nsub is the number of subspace variables in free set. c On exit nsub is not changed. c c ind is an integer array of dimension nsub. c On entry ind specifies the indices of subspace variables. c On exit ind is unchanged. c c nenter is an integer variable. c On entry nenter is the number of variables entering the c free set. c On exit nenter is unchanged. c c ileave is an integer variable. c On entry indx2(ileave),...,indx2(n) are the variables leaving c the free set. c On exit ileave is unchanged. c c indx2 is an integer array of dimension n. c On entry indx2(1),...,indx2(nenter) are the variables entering c the free set, while indx2(ileave),...,indx2(n) are the c variables leaving the free set. c On exit indx2 is unchanged. c c iupdat is an integer variable. c On entry iupdat is the total number of BFGS updates made so far. c On exit iupdat is unchanged. c c updatd is a logical variable. c On entry 'updatd' is true if the L-BFGS matrix is updatd. c On exit 'updatd' is unchanged. c c wn is a double precision array of dimension 2m x 2m. c On entry wn is unspecified. c On exit the upper triangle of wn stores the LEL^T factorization c of the 2*col x 2*col indefinite matrix c [-D -Y'ZZ'Y/theta L_a'-R_z' ] c [L_a -R_z theta*S'AA'S ] c c wn1 is a double precision array of dimension 2m x 2m. c On entry wn1 stores the lower triangular part of c [Y' ZZ'Y L_a'+R_z'] c [L_a+R_z S'AA'S ] c in the previous iteration. c On exit wn1 stores the corresponding updated matrices. c The purpose of wn1 is just to store these inner products c so they can be easily updated and inserted into wn. c c m is an integer variable. c On entry m is the maximum number of variable metric corrections c used to define the limited memory matrix. c On exit m is unchanged. c c ws, wy, sy, and wtyy are double precision arrays; c theta is a double precision variable; c col is an integer variable; c head is an integer variable. c On entry they store the information defining the c limited memory BFGS matrix: c ws(n,m) stores S, a set of s-vectors; c wy(n,m) stores Y, a set of y-vectors; c sy(m,m) stores S'Y; c wtyy(m,m) stores the Cholesky factorization c of (theta*S'S+LD^(-1)L') c theta is the scaling factor specifying B_0 = theta I; c col is the number of variable metric corrections stored; c head is the location of the 1st s- (or y-) vector in S (or Y). c On exit they are unchanged. c c info is an integer variable. c On entry info is unspecified. c On exit info = 0 for normal return; c = -1 when the 1st Cholesky factorization failed; c = -2 when the 2st Cholesky factorization failed. c c Subprograms called: c c Linpack ... dcopy, dpofa, dtrsl. c c c References: c [1] R. H. Byrd, P. Lu, J. Nocedal and C. Zhu, ``A limited c memory algorithm for bound constrained optimization'', c SIAM J. Scientific Computing 16 (1995), no. 5, pp. 1190--1208. c c [2] C. Zhu, R.H. Byrd, P. Lu, J. Nocedal, ``L-BFGS-B: a c limited memory FORTRAN code for solving bound constrained c optimization problems'', Tech. Report, NAM-11, EECS Department, c Northwestern University, 1994. c c (Postscript files of these papers are available via anonymous c ftp to ece.nwu.edu in the directory pub/lbfgs/lbfgs_bcm.) c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer m2,ipntr,jpntr,iy,is,jy,js,is1,js1,k1,i,k, + col2,pbegin,pend,dbegin,dend,upcl double precision ddot,temp1,temp2,temp3,temp4 double precision zero parameter (zero=0.0d0) c Form the lower triangular part of c WN1 = [Y' ZZ'Y L_a'+R_z'] c [L_a+R_z S'AA'S ] c where L_a is the strictly lower triangular part of S'AA'Y c R_z is the upper triangular part of S'ZZ'Y. if (updatd) then if (iupdat .gt. m) then c shift old part of WN1. do 10 jy = 1, m - 1 js = m + jy call dcopy(m-jy,wn1(jy+1,jy+1),1,wn1(jy,jy),1) call dcopy(m-jy,wn1(js+1,js+1),1,wn1(js,js),1) call dcopy(m-1,wn1(m+2,jy+1),1,wn1(m+1,jy),1) 10 continue endif c put new rows in blocks (1,1), (2,1) and (2,2). pbegin = 1 pend = nsub dbegin = nsub + 1 dend = n iy = col is = m + col ipntr = head + col - 1 if (ipntr .gt. m) ipntr = ipntr - m jpntr = head do 20 jy = 1, col js = m + jy temp1 = zero temp2 = zero temp3 = zero c compute element jy of row 'col' of Y'ZZ'Y do 15 k = pbegin, pend k1 = ind(k) temp1 = temp1 + wy(k1,ipntr)*wy(k1,jpntr) 15 continue c compute elements jy of row 'col' of L_a and S'AA'S do 16 k = dbegin, dend k1 = ind(k) temp2 = temp2 + ws(k1,ipntr)*ws(k1,jpntr) temp3 = temp3 + ws(k1,ipntr)*wy(k1,jpntr) 16 continue wn1(iy,jy) = temp1 wn1(is,js) = temp2 wn1(is,jy) = temp3 jpntr = mod(jpntr,m) + 1 20 continue c put new column in block (2,1). jy = col jpntr = head + col - 1 if (jpntr .gt. m) jpntr = jpntr - m ipntr = head do 30 i = 1, col is = m + i temp3 = zero c compute element i of column 'col' of R_z do 25 k = pbegin, pend k1 = ind(k) temp3 = temp3 + ws(k1,ipntr)*wy(k1,jpntr) 25 continue ipntr = mod(ipntr,m) + 1 wn1(is,jy) = temp3 30 continue upcl = col - 1 else upcl = col endif c modify the old parts in blocks (1,1) and (2,2) due to changes c in the set of free variables. ipntr = head do 45 iy = 1, upcl is = m + iy jpntr = head do 40 jy = 1, iy js = m + jy temp1 = zero temp2 = zero temp3 = zero temp4 = zero do 35 k = 1, nenter k1 = indx2(k) temp1 = temp1 + wy(k1,ipntr)*wy(k1,jpntr) temp2 = temp2 + ws(k1,ipntr)*ws(k1,jpntr) 35 continue do 36 k = ileave, n k1 = indx2(k) temp3 = temp3 + wy(k1,ipntr)*wy(k1,jpntr) temp4 = temp4 + ws(k1,ipntr)*ws(k1,jpntr) 36 continue wn1(iy,jy) = wn1(iy,jy) + temp1 - temp3 wn1(is,js) = wn1(is,js) - temp2 + temp4 jpntr = mod(jpntr,m) + 1 40 continue ipntr = mod(ipntr,m) + 1 45 continue c modify the old parts in block (2,1). ipntr = head do 60 is = m + 1, m + upcl jpntr = head do 55 jy = 1, upcl temp1 = zero temp3 = zero do 50 k = 1, nenter k1 = indx2(k) temp1 = temp1 + ws(k1,ipntr)*wy(k1,jpntr) 50 continue do 51 k = ileave, n k1 = indx2(k) temp3 = temp3 + ws(k1,ipntr)*wy(k1,jpntr) 51 continue if (is .le. jy + m) then wn1(is,jy) = wn1(is,jy) + temp1 - temp3 else wn1(is,jy) = wn1(is,jy) - temp1 + temp3 endif jpntr = mod(jpntr,m) + 1 55 continue ipntr = mod(ipntr,m) + 1 60 continue c Form the upper triangle of WN = [D+Y' ZZ'Y/theta -L_a'+R_z' ] c [-L_a +R_z S'AA'S*theta] m2 = 2*m do 70 iy = 1, col is = col + iy is1 = m + iy do 65 jy = 1, iy js = col + jy js1 = m + jy wn(jy,iy) = wn1(iy,jy)/theta wn(js,is) = wn1(is1,js1)*theta 65 continue do 66 jy = 1, iy - 1 wn(jy,is) = -wn1(is1,jy) 66 continue do 67 jy = iy, col wn(jy,is) = wn1(is1,jy) 67 continue wn(iy,iy) = wn(iy,iy) + sy(iy,iy) 70 continue c Form the upper triangle of c WN= [ LL' L^-1(-L_a'+R_z')] c [(-L_a +R_z)L'^-1 S'AA'S*theta ] c first Cholesky factor (1,1) block of wn to get LL' c with L' stored in the upper triangle of wn. call dpofa(wn,m2,col,info) if (info .ne. 0) then info = -1 return endif c then form L^-1(-L_a'+R_z') in the (1,2) block. col2 = 2*col do 71 js = col+1 ,col2 call dtrsl(wn,m2,col,wn(1,js),11,info) 71 continue c Form S'AA'S*theta + (L^-1(-L_a'+R_z'))'L^-1(-L_a'+R_z') in the c upper triangle of (2,2) block of wn. do 72 is = col+1, col2 do 74 js = is, col2 wn(is,js) = wn(is,js) + ddot(col,wn(1,is),1,wn(1,js),1) 74 continue 72 continue c Cholesky factorization of (2,2) block of wn. call dpofa(wn(col+1,col+1),m2,col,info) if (info .ne. 0) then info = -2 return endif return end c======================= The end of formk ============================== subroutine formt(m, wt, sy, ss, col, theta, info) integer m, col, info double precision theta, wt(m, m), sy(m, m), ss(m, m) c ************ c c Subroutine formt c c This subroutine forms the upper half of the pos. def. and symm. c T = theta*SS + L*D^(-1)*L', stores T in the upper triangle c of the array wt, and performs the Cholesky factorization of T c to produce J*J', with J' stored in the upper triangle of wt. c c Subprograms called: c c Linpack ... dpofa. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i,j,k,k1 double precision ddum double precision zero parameter (zero=0.0d0) c Form the upper half of T = theta*SS + L*D^(-1)*L', c store T in the upper triangle of the array wt. do 52 j = 1, col wt(1,j) = theta*ss(1,j) 52 continue do 55 i = 2, col do 54 j = i, col k1 = min(i,j) - 1 ddum = zero do 53 k = 1, k1 ddum = ddum + sy(i,k)*sy(j,k)/sy(k,k) 53 continue wt(i,j) = ddum + theta*ss(i,j) 54 continue 55 continue c Cholesky factorize T to J*J' with c J' stored in the upper triangle of wt. call dpofa(wt,m,col,info) if (info .ne. 0) then info = -3 endif return end c======================= The end of formt ============================== subroutine freev(n, nfree, index, nenter, ileave, indx2, + iwhere, wrk, updatd, cnstnd, iprint, iter) integer n, nfree, nenter, ileave, iprint, iter, + index(n), indx2(n), iwhere(n) logical wrk, updatd, cnstnd c ************ c c Subroutine freev c c This subroutine counts the entering and leaving variables when c iter > 0, and finds the index set of free and active variables c at the GCP. c c cnstnd is a logical variable indicating whether bounds are present c c index is an integer array of dimension n c for i=1,...,nfree, index(i) are the indices of free variables c for i=nfree+1,...,n, index(i) are the indices of bound variables c On entry after the first iteration, index gives c the free variables at the previous iteration. c On exit it gives the free variables based on the determination c in cauchy using the array iwhere. c c indx2 is an integer array of dimension n c On entry indx2 is unspecified. c On exit with iter>0, indx2 indicates which variables c have changed status since the previous iteration. c For i= 1,...,nenter, indx2(i) have changed from bound to free. c For i= ileave+1,...,n, indx2(i) have changed from free to bound. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer iact,i,k nenter = 0 ileave = n + 1 if (iter .gt. 0 .and. cnstnd) then c count the entering and leaving variables. do 20 i = 1, nfree k = index(i) if (iwhere(k) .gt. 0) then ileave = ileave - 1 indx2(ileave) = k if (iprint .ge. 100) write (6,*) + 'Variable ',k,' leaves the set of free variables' endif 20 continue do 22 i = 1 + nfree, n k = index(i) if (iwhere(k) .le. 0) then nenter = nenter + 1 indx2(nenter) = k if (iprint .ge. 100) write (6,*) + 'Variable ',k,' enters the set of free variables' endif 22 continue if (iprint .ge. 99) write (6,*) + n+1-ileave,' variables leave; ',nenter,' variables enter' endif wrk = (ileave .lt. n+1) .or. (nenter .gt. 0) .or. updatd c Find the index set of free and active variables at the GCP. nfree = 0 iact = n + 1 do 24 i = 1, n if (iwhere(i) .le. 0) then nfree = nfree + 1 index(nfree) = i else iact = iact - 1 index(iact) = i endif 24 continue if (iprint .ge. 99) write (6,*) + nfree,' variables are free at GCP ',iter + 1 return end c======================= The end of freev ============================== subroutine hpsolb(n, t, iorder, iheap) integer iheap, n, iorder(n) double precision t(n) c ************ c c Subroutine hpsolb c c This subroutine sorts out the least element of t, and puts the c remaining elements of t in a heap. c c n is an integer variable. c On entry n is the dimension of the arrays t and iorder. c On exit n is unchanged. c c t is a double precision array of dimension n. c On entry t stores the elements to be sorted, c On exit t(n) stores the least elements of t, and t(1) to t(n-1) c stores the remaining elements in the form of a heap. c c iorder is an integer array of dimension n. c On entry iorder(i) is the index of t(i). c On exit iorder(i) is still the index of t(i), but iorder may be c permuted in accordance with t. c c iheap is an integer variable specifying the task. c On entry iheap should be set as follows: c iheap .eq. 0 if t(1) to t(n) is not in the form of a heap, c iheap .ne. 0 if otherwise. c On exit iheap is unchanged. c c c References: c Algorithm 232 of CACM (J. W. J. Williams): HEAPSORT. c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c ************ integer i,j,k,indxin,indxou double precision ddum,out if (iheap .eq. 0) then c Rearrange the elements t(1) to t(n) to form a heap. do 20 k = 2, n ddum = t(k) indxin = iorder(k) c Add ddum to the heap. i = k 10 continue if (i.gt.1) then j = i/2 if (ddum .lt. t(j)) then t(i) = t(j) iorder(i) = iorder(j) i = j goto 10 endif endif t(i) = ddum iorder(i) = indxin 20 continue endif c Assign to 'out' the value of t(1), the least member of the heap, c and rearrange the remaining members to form a heap as c elements 1 to n-1 of t. if (n .gt. 1) then i = 1 out = t(1) indxou = iorder(1) ddum = t(n) indxin = iorder(n) c Restore the heap 30 continue j = i+i if (j .le. n-1) then if (t(j+1) .lt. t(j)) j = j+1 if (t(j) .lt. ddum ) then t(i) = t(j) iorder(i) = iorder(j) i = j goto 30 endif endif t(i) = ddum iorder(i) = indxin c Put the least member in t(n). t(n) = out iorder(n) = indxou endif return end c====================== The end of hpsolb ============================== subroutine lnsrlb(n, l, u, nbd, x, f, fold, gd, gdold, g, d, r, t, + z, stp, dnorm, dtd, xstep, stpmx, iter, ifun, + iback, nfgv, info, task, boxed, cnstnd, csave, + isave, dsave) character*60 task, csave logical boxed, cnstnd integer n, iter, ifun, iback, nfgv, info, + nbd(n), isave(2) double precision f, fold, gd, gdold, stp, dnorm, dtd, xstep, + stpmx, x(n), l(n), u(n), g(n), d(n), r(n), t(n), + z(n), dsave(13) c ********** c c Subroutine lnsrlb c c This subroutine calls subroutine dcsrch from the Minpack2 library c to perform the line search. Subroutine dscrch is safeguarded so c that all trial points lie within the feasible region. c c Subprograms called: c c Minpack2 Library ... dcsrch. c c Linpack ... dtrsl, ddot. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ********** integer i double precision ddot,a1,a2 double precision one,zero,big parameter (one=1.0d0,zero=0.0d0,big=1.0d+10) double precision ftol,gtol,xtol parameter (ftol=1.0d-3,gtol=0.9d0,xtol=0.1d0) if (task(1:5) .eq. 'FG_LN') goto 556 dtd = ddot(n,d,1,d,1) dnorm = sqrt(dtd) c Determine the maximum step length. stpmx = big if (cnstnd) then if (iter .eq. 0) then stpmx = one else do 43 i = 1, n a1 = d(i) if (nbd(i) .ne. 0) then if (a1 .lt. zero .and. nbd(i) .le. 2) then a2 = l(i) - x(i) if (a2 .ge. zero) then stpmx = zero else if (a1*stpmx .lt. a2) then stpmx = a2/a1 endif else if (a1 .gt. zero .and. nbd(i) .ge. 2) then a2 = u(i) - x(i) if (a2 .le. zero) then stpmx = zero else if (a1*stpmx .gt. a2) then stpmx = a2/a1 endif endif endif 43 continue endif endif if (iter .eq. 0 .and. .not. boxed) then stp = min(one/dnorm, stpmx) else stp = one endif call dcopy(n,x,1,t,1) call dcopy(n,g,1,r,1) fold = f ifun = 0 iback = 0 csave = 'START' 556 continue gd = ddot(n,g,1,d,1) if (ifun .eq. 0) then gdold=gd if (gd .ge. zero) then c the directional derivative >=0. c Line search is impossible. info = -4 return endif endif call dcsrch(f,gd,stp,ftol,gtol,xtol,zero,stpmx,csave,isave,dsave) xstep = stp*dnorm if (csave(1:4) .ne. 'CONV' .and. csave(1:4) .ne. 'WARN') then task = 'FG_LNSRCH' ifun = ifun + 1 nfgv = nfgv + 1 iback = ifun - 1 if (stp .eq. one) then call dcopy(n,z,1,x,1) else do 41 i = 1, n x(i) = stp*d(i) + t(i) 41 continue endif else task = 'NEW_X' endif return end c======================= The end of lnsrlb ============================= subroutine matupd(n, m, ws, wy, sy, ss, d, r, itail, + iupdat, col, head, theta, rr, dr, stp, dtd) integer n, m, itail, iupdat, col, head double precision theta, rr, dr, stp, dtd, d(n), r(n), + ws(n, m), wy(n, m), sy(m, m), ss(m, m) c ************ c c Subroutine matupd c c This subroutine updates matrices WS and WY, and forms the c middle matrix in B. c c Subprograms called: c c Linpack ... dcopy, ddot. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer j,pointr double precision ddot double precision one parameter (one=1.0d0) c Set pointers for matrices WS and WY. if (iupdat .le. m) then col = iupdat itail = mod(head+iupdat-2,m) + 1 else itail = mod(itail,m) + 1 head = mod(head,m) + 1 endif c Update matrices WS and WY. call dcopy(n,d,1,ws(1,itail),1) call dcopy(n,r,1,wy(1,itail),1) c Set theta=yy/ys. theta = rr/dr c Form the middle matrix in B. c update the upper triangle of SS, c and the lower triangle of SY: if (iupdat .gt. m) then c move old information do 50 j = 1, col - 1 call dcopy(j,ss(2,j+1),1,ss(1,j),1) call dcopy(col-j,sy(j+1,j+1),1,sy(j,j),1) 50 continue endif c add new information: the last row of SY c and the last column of SS: pointr = head do 51 j = 1, col - 1 sy(col,j) = ddot(n,d,1,wy(1,pointr),1) ss(j,col) = ddot(n,ws(1,pointr),1,d,1) pointr = mod(pointr,m) + 1 51 continue if (stp .eq. one) then ss(col,col) = dtd else ss(col,col) = stp*stp*dtd endif sy(col,col) = dr return end c======================= The end of matupd ============================= subroutine prn1lb(n, m, l, u, x, iprint, itfile, epsmch) integer n, m, iprint, itfile double precision epsmch, x(n), l(n), u(n) c ************ c c Subroutine prn1lb c c This subroutine prints the input data, initial point, upper and c lower bounds of each variable, machine precision, as well as c the headings of the output. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i if (iprint .ge. 0) then write (6,7001) epsmch write (6,*) 'N = ',n,' M = ',m if (iprint .ge. 1) then write (itfile,2001) epsmch write (itfile,*)'N = ',n,' M = ',m write (itfile,9001) if (iprint .gt. 100) then write (6,1004) 'L =',(l(i),i = 1,n) write (6,1004) 'X0 =',(x(i),i = 1,n) write (6,1004) 'U =',(u(i),i = 1,n) endif endif endif 1004 format (/,a4, 1p, 6(1x,d11.4),/,(4x,1p,6(1x,d11.4))) 2001 format ('RUNNING THE L-BFGS-B CODE',/,/, + 'it = iteration number',/, + 'nf = number of function evaluations',/, + 'nint = number of segments explored during the Cauchy search',/, + 'nact = number of active bounds at the generalized Cauchy point' + ,/, + 'sub = manner in which the subspace minimization terminated:' + ,/,' con = converged, bnd = a bound was reached',/, + 'itls = number of iterations performed in the line search',/, + 'stepl = step length used',/, + 'tstep = norm of the displacement (total step)',/, + 'projg = norm of the projected gradient',/, + 'f = function value',/,/, + ' * * *',/,/, + 'Machine precision =',1p,d10.3) 7001 format ('RUNNING THE L-BFGS-B CODE',/,/, + ' * * *',/,/, + 'Machine precision =',1p,d10.3) 9001 format (/,3x,'it',3x,'nf',2x,'nint',2x,'nact',2x,'sub',2x,'itls', + 2x,'stepl',4x,'tstep',5x,'projg',8x,'f') return end c======================= The end of prn1lb ============================= subroutine prn2lb(n, x, f, g, iprint, itfile, iter, nfgv, nact, + sbgnrm, nint, word, iword, iback, stp, xstep) character*3 word integer n, iprint, itfile, iter, nfgv, nact, nint, + iword, iback double precision f, sbgnrm, stp, xstep, x(n), g(n) c ************ c c Subroutine prn2lb c c This subroutine prints out new information after a successful c line search. c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i,imod c 'word' records the status of subspace solutions. if (iword .eq. 0) then c the subspace minimization converged. word = 'con' else if (iword .eq. 1) then c the subspace minimization stopped at a bound. word = 'bnd' else if (iword .eq. 5) then c the truncated Newton step has been used. word = 'TNT' else word = '---' endif if (iprint .ge. 99) then write (6,*) 'LINE SEARCH',iback,' times; norm of step = ',xstep write (6,2001) iter,f,sbgnrm if (iprint .gt. 100) then write (6,1004) 'X =',(x(i), i = 1, n) write (6,1004) 'G =',(g(i), i = 1, n) endif else if (iprint .gt. 0) then imod = mod(iter,iprint) if (imod .eq. 0) write (6,2001) iter,f,sbgnrm endif if (iprint .ge. 1) write (itfile,3001) + iter,nfgv,nint,nact,word,iback,stp,xstep,sbgnrm,f 1004 format (/,a4, 1p, 6(1x,d11.4),/,(4x,1p,6(1x,d11.4))) 2001 format + (/,'At iterate',i5,4x,'f= ',1p,d12.5,4x,'|proj g|= ',1p,d12.5) 3001 format(2(1x,i4),2(1x,i5),2x,a3,1x,i4,1p,2(2x,d7.1),1p,2(1x,d10.3)) return end c======================= The end of prn2lb ============================= subroutine prn3lb(n, x, f, task, iprint, info, itfile, + iter, nfgv, nintol, nskip, nact, sbgnrm, + time, nint, word, iback, stp, xstep, k, + cachyt, sbtime, lnscht) character*60 task character*3 word integer n, iprint, info, itfile, iter, nfgv, nintol, + nskip, nact, nint, iback, k double precision f, sbgnrm, time, stp, xstep, cachyt, sbtime, + lnscht, x(n) c ************ c c Subroutine prn3lb c c This subroutine prints out information when either a built-in c convergence test is satisfied or when an error message is c generated. c c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i if (task(1:5) .eq. 'ERROR') goto 999 if (iprint .ge. 0) then write (6,3003) write (6,3004) write(6,3005) n,iter,nfgv,nintol,nskip,nact,sbgnrm,f if (iprint .ge. 100) then write (6,1004) 'X =',(x(i),i = 1,n) endif if (iprint .ge. 1) write (6,*) ' F =',f endif 999 continue if (iprint .ge. 0) then write (6,3009) task if (info .ne. 0) then if (info .eq. -1) write (6,9011) if (info .eq. -2) write (6,9012) if (info .eq. -3) write (6,9013) if (info .eq. -4) write (6,9014) if (info .eq. -5) write (6,9015) if (info .eq. -6) write (6,*)' Input nbd(',k,') is invalid.' if (info .eq. -7) + write (6,*)' l(',k,') > u(',k,'). No feasible solution.' if (info .eq. -8) write (6,9018) if (info .eq. -9) write (6,9019) endif if (iprint .ge. 1) write (6,3007) cachyt,sbtime,lnscht write (6,3008) time if (iprint .ge. 1) then if (info .eq. -4 .or. info .eq. -9) then write (itfile,3002) + iter,nfgv,nint,nact,word,iback,stp,xstep endif write (itfile,3009) task if (info .ne. 0) then if (info .eq. -1) write (itfile,9011) if (info .eq. -2) write (itfile,9012) if (info .eq. -3) write (itfile,9013) if (info .eq. -4) write (itfile,9014) if (info .eq. -5) write (itfile,9015) if (info .eq. -8) write (itfile,9018) if (info .eq. -9) write (itfile,9019) endif write (itfile,3008) time endif endif 1004 format (/,a4, 1p, 6(1x,d11.4),/,(4x,1p,6(1x,d11.4))) 3002 format(2(1x,i4),2(1x,i5),2x,a3,1x,i4,1p,2(2x,d7.1),6x,'-',10x,'-') 3003 format (/, + ' * * *',/,/, + 'Tit = total number of iterations',/, + 'Tnf = total number of function evaluations',/, + 'Tnint = total number of segments explored during', + ' Cauchy searches',/, + 'Skip = number of BFGS updates skipped',/, + 'Nact = number of active bounds at final generalized', + ' Cauchy point',/, + 'Projg = norm of the final projected gradient',/, + 'F = final function value',/,/, + ' * * *') 3004 format (/,3x,'N',3x,'Tit',2x,'Tnf',2x,'Tnint',2x, + 'Skip',2x,'Nact',5x,'Projg',8x,'F') 3005 format (i5,2(1x,i4),(1x,i6),(2x,i4),(1x,i5),1p,2(2x,d10.3)) 3007 format (/,' Cauchy time',1p,e10.3,' seconds.',/ + ' Subspace minimization time',1p,e10.3,' seconds.',/ + ' Line search time',1p,e10.3,' seconds.') 3008 format (/,' Total User time',1p,e10.3,' seconds.',/) 3009 format (/,a60) 9011 format (/, +' Matrix in 1st Cholesky factorization in formk is not Pos. Def.') 9012 format (/, +' Matrix in 2st Cholesky factorization in formk is not Pos. Def.') 9013 format (/, +' Matrix in the Cholesky factorization in formt is not Pos. Def.') 9014 format (/, +' Derivative >= 0, backtracking line search impossible.',/, +' Previous x, f and g restored.',/, +' Possible causes: 1 error in function or gradient evaluation;',/, +' 2 rounding errors dominate computation.') 9015 format (/, +' Warning: more than 10 function and gradient',/, +' evaluations in the last line search. Termination',/, +' may possibly be caused by a bad search direction.') 9018 format (/,' The triangular system is singular.') 9019 format (/, +' Line search cannot locate an adequate point after 20 function',/ +,' and gradient evaluations. Previous x, f and g restored.',/, +' Possible causes: 1 error in function or gradient evaluation;',/, +' 2 rounding error dominate computation.') return end c======================= The end of prn3lb ============================= subroutine projgr(n, l, u, nbd, x, g, sbgnrm) integer n, nbd(n) double precision sbgnrm, x(n), l(n), u(n), g(n) c ************ c c Subroutine projgr c c This subroutine computes the infinity norm of the projected c gradient. c c c * * * c c NEOS, November 1994. (Latest revision April 1997.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ integer i double precision gi double precision zero parameter (zero=0.0d0) sbgnrm = zero do 15 i = 1, n gi = g(i) if (nbd(i) .ne. 0) then if (gi .lt. zero) then if (nbd(i) .ge. 2) gi = max((x(i)-u(i)),gi) else if (nbd(i) .le. 2) gi = min((x(i)-l(i)),gi) endif endif sbgnrm = max(sbgnrm,abs(gi)) 15 continue return end c======================= The end of projgr ============================= subroutine subsm(n, m, nsub, ind, l, u, nbd, x, d, ws, wy, theta, + col, head, iword, wv, wn, iprint, info) integer n, m, nsub, col, head, iword, iprint, info, + ind(nsub), nbd(n) double precision theta, + l(n), u(n), x(n), d(n), + ws(n, m), wy(n, m), + wv(2*m), wn(2*m, 2*m) c ************ c c Subroutine subsm c c Given xcp, l, u, r, an index set that specifies c the active set at xcp, and an l-BFGS matrix B c (in terms of WY, WS, SY, WT, head, col, and theta), c this subroutine computes an approximate solution c of the subspace problem c c (P) min Q(x) = r'(x-xcp) + 1/2 (x-xcp)' B (x-xcp) c c subject to l<=x<=u c x_i=xcp_i for all i in A(xcp) c c along the subspace unconstrained Newton direction c c d = -(Z'BZ)^(-1) r. c c The formula for the Newton direction, given the L-BFGS matrix c and the Sherman-Morrison formula, is c c d = (1/theta)r + (1/theta*2) Z'WK^(-1)W'Z r. c c where c K = [-D -Y'ZZ'Y/theta L_a'-R_z' ] c [L_a -R_z theta*S'AA'S ] c c Note that this procedure for computing d differs c from that described in [1]. One can show that the matrix K is c equal to the matrix M^[-1]N in that paper. c c n is an integer variable. c On entry n is the dimension of the problem. c On exit n is unchanged. c c m is an integer variable. c On entry m is the maximum number of variable metric corrections c used to define the limited memory matrix. c On exit m is unchanged. c c nsub is an integer variable. c On entry nsub is the number of free variables. c On exit nsub is unchanged. c c ind is an integer array of dimension nsub. c On entry ind specifies the coordinate indices of free variables. c On exit ind is unchanged. c c l is a double precision array of dimension n. c On entry l is the lower bound of x. c On exit l is unchanged. c c u is a double precision array of dimension n. c On entry u is the upper bound of x. c On exit u is unchanged. c c nbd is a integer array of dimension n. c On entry nbd represents the type of bounds imposed on the c variables, and must be specified as follows: c nbd(i)=0 if x(i) is unbounded, c 1 if x(i) has only a lower bound, c 2 if x(i) has both lower and upper bounds, and c 3 if x(i) has only an upper bound. c On exit nbd is unchanged. c c x is a double precision array of dimension n. c On entry x specifies the Cauchy point xcp. c On exit x(i) is the minimizer of Q over the subspace of c free variables. c c d is a double precision array of dimension n. c On entry d is the reduced gradient of Q at xcp. c On exit d is the Newton direction of Q. c c ws and wy are double precision arrays; c theta is a double precision variable; c col is an integer variable; c head is an integer variable. c On entry they store the information defining the c limited memory BFGS matrix: c ws(n,m) stores S, a set of s-vectors; c wy(n,m) stores Y, a set of y-vectors; c theta is the scaling factor specifying B_0 = theta I; c col is the number of variable metric corrections stored; c head is the location of the 1st s- (or y-) vector in S (or Y). c On exit they are unchanged. c c iword is an integer variable. c On entry iword is unspecified. c On exit iword specifies the status of the subspace solution. c iword = 0 if the solution is in the box, c 1 if some bound is encountered. c c wv is a double precision working array of dimension 2m. c c wn is a double precision array of dimension 2m x 2m. c On entry the upper triangle of wn stores the LEL^T factorization c of the indefinite matrix c c K = [-D -Y'ZZ'Y/theta L_a'-R_z' ] c [L_a -R_z theta*S'AA'S ] c where E = [-I 0] c [ 0 I] c On exit wn is unchanged. c c iprint is an INTEGER variable that must be set by the user. c It controls the frequency and type of output generated: c iprint<0 no output is generated; c iprint=0 print only one line at the last iteration; c 0100 print details of every iteration including x and g; c When iprint > 0, the file iterate.dat will be created to c summarize the iteration. c c info is an integer variable. c On entry info is unspecified. c On exit info = 0 for normal return, c = nonzero for abnormal return c when the matrix K is ill-conditioned. c c Subprograms called: c c Linpack dtrsl. c c c References: c c [1] R. H. Byrd, P. Lu, J. Nocedal and C. Zhu, ``A limited c memory algorithm for bound constrained optimization'', c SIAM J. Scientific Computing 16 (1995), no. 5, pp. 1190--1208. c c c c * * * c c NEOS, November 1994. (Latest revision June 1996.) c Optimization Technology Center. c Argonne National Laboratory and Northwestern University. c Written by c Ciyou Zhu c in collaboration with R.H. Byrd, P. Lu-Chen and J. Nocedal. c c c ************ logical temp1_updated integer pointr,m2,col2,ibd,jy,js,i,j,k double precision alpha,dk,temp1,temp2 double precision one,zero parameter (one=1.0d0,zero=0.0d0) if (nsub .le. 0) return if (iprint .ge. 99) write (6,1001) c Compute wv = W'Zd. pointr = head do 20 i = 1, col temp1 = zero temp2 = zero do 10 j = 1, nsub k = ind(j) temp1 = temp1 + wy(k,pointr)*d(j) temp2 = temp2 + ws(k,pointr)*d(j) 10 continue wv(i) = temp1 wv(col + i) = theta*temp2 pointr = mod(pointr,m) + 1 20 continue c Compute wv:=K^(-1)wv. m2 = 2*m col2 = 2*col call dtrsl(wn,m2,col2,wv,11,info) if (info .ne. 0) return do 25 i = 1, col wv(i) = -wv(i) 25 continue call dtrsl(wn,m2,col2,wv,01,info) if (info .ne. 0) return c Compute d = (1/theta)d + (1/theta**2)Z'W wv. pointr = head do 40 jy = 1, col js = col + jy do 30 i = 1, nsub k = ind(i) d(i) = d(i) + wy(k,pointr)*wv(jy)/theta + + ws(k,pointr)*wv(js) 30 continue pointr = mod(pointr,m) + 1 40 continue do 50 i = 1, nsub d(i) = d(i)/theta 50 continue c Backtrack to the feasible region. alpha = one temp1 = alpha do 60 i = 1, nsub k = ind(i) dk = d(i) if (nbd(k) .ne. 0) then temp1_updated = .false. if (dk .lt. zero .and. nbd(k) .le. 2) then temp2 = l(k) - x(k) if (temp2 .ge. zero) then temp1 = zero temp1_updated = .true. else if (dk*alpha .lt. temp2) then temp1 = temp2/dk temp1_updated = .true. endif else if (dk .gt. zero .and. nbd(k) .ge. 2) then temp2 = u(k) - x(k) if (temp2 .le. zero) then temp1 = zero temp1_updated = .true. else if (dk*alpha .gt. temp2) then temp1 = temp2/dk temp1_updated = .true. endif endif cc logical variable temp1_updated added to eliminate unexpected cc trigger of the if statement due to possible difference between cc hardware precision and double precision. if (temp1_updated .and. temp1 .lt. alpha) then alpha = temp1 ibd = i endif endif 60 continue if (alpha .lt. one) then dk = d(ibd) k = ind(ibd) if (dk .gt. zero) then x(k) = u(k) d(ibd) = zero else if (dk .lt. zero) then x(k) = l(k) d(ibd) = zero endif endif do 70 i = 1, nsub k = ind(i) x(k) = x(k) + alpha*d(i) 70 continue if (iprint .ge. 99) then if (alpha .lt. one) then write (6,1002) alpha else write (6,*) 'SM solution inside the box' end if if (iprint .gt.100) write (6,1003) (x(i),i=1,n) endif if (alpha .lt. one) then iword = 1 else iword = 0 endif if (iprint .ge. 99) write (6,1004) 1001 format (/,'----------------SUBSM entered-----------------',/) 1002 format ( 'ALPHA = ',f7.5,' backtrack to the BOX') 1003 format ('Subspace solution X = ',/,(4x,1p,6(1x,d11.4))) 1004 format (/,'----------------exit SUBSM --------------------',/) return end c====================== The end of subsm =============================== subroutine dcsrch(f,g,stp,ftol,gtol,xtol,stpmin,stpmax, + task,isave,dsave) character*(*) task integer isave(2) double precision f,g,stp,ftol,gtol,xtol,stpmin,stpmax double precision dsave(13) c ********** c c Subroutine dcsrch c c This subroutine finds a step that satisfies a sufficient c decrease condition and a curvature condition. c c Each call of the subroutine updates an interval with c endpoints stx and sty. The interval is initially chosen c so that it contains a minimizer of the modified function c c psi(stp) = f(stp) - f(0) - ftol*stp*f'(0). c c If psi(stp) <= 0 and f'(stp) >= 0 for some step, then the c interval is chosen so that it contains a minimizer of f. c c The algorithm is designed to find a step that satisfies c the sufficient decrease condition c c f(stp) <= f(0) + ftol*stp*f'(0), c c and the curvature condition c c abs(f'(stp)) <= gtol*abs(f'(0)). c c If ftol is less than gtol and if, for example, the function c is bounded below, then there is always a step which satisfies c both conditions. c c If no step can be found that satisfies both conditions, then c the algorithm stops with a warning. In this case stp only c satisfies the sufficient decrease condition. c c A typical invocation of dcsrch has the following outline: c c task = 'START' c 10 continue c call dcsrch( ... ) c if (task .eq. 'FG') then c Evaluate the function and the gradient at stp c goto 10 c end if c c NOTE: The user must no alter work arrays between calls. c c The subroutine statement is c c subroutine dcsrch(f,g,stp,ftol,gtol,xtol,stpmin,stpmax, c task,isave,dsave) c where c c f is a double precision variable. c On initial entry f is the value of the function at 0. c On subsequent entries f is the value of the c function at stp. c On exit f is the value of the function at stp. c c g is a double precision variable. c On initial entry g is the derivative of the function at 0. c On subsequent entries g is the derivative of the c function at stp. c On exit g is the derivative of the function at stp. c c stp is a double precision variable. c On entry stp is the current estimate of a satisfactory c step. On initial entry, a positive initial estimate c must be provided. c On exit stp is the current estimate of a satisfactory step c if task = 'FG'. If task = 'CONV' then stp satisfies c the sufficient decrease and curvature condition. c c ftol is a double precision variable. c On entry ftol specifies a nonnegative tolerance for the c sufficient decrease condition. c On exit ftol is unchanged. c c gtol is a double precision variable. c On entry gtol specifies a nonnegative tolerance for the c curvature condition. c On exit gtol is unchanged. c c xtol is a double precision variable. c On entry xtol specifies a nonnegative relative tolerance c for an acceptable step. The subroutine exits with a c warning if the relative difference between sty and stx c is less than xtol. c On exit xtol is unchanged. c c stpmin is a double precision variable. c On entry stpmin is a nonnegative lower bound for the step. c On exit stpmin is unchanged. c c stpmax is a double precision variable. c On entry stpmax is a nonnegative upper bound for the step. c On exit stpmax is unchanged. c c task is a character variable of length at least 60. c On initial entry task must be set to 'START'. c On exit task indicates the required action: c c If task(1:2) = 'FG' then evaluate the function and c derivative at stp and call dcsrch again. c c If task(1:4) = 'CONV' then the search is successful. c c If task(1:4) = 'WARN' then the subroutine is not able c to satisfy the convergence conditions. The exit value of c stp contains the best point found during the search. c c If task(1:5) = 'ERROR' then there is an error in the c input arguments. c c On exit with convergence, a warning or an error, the c variable task contains additional information. c c isave is an integer work array of dimension 2. c c dsave is a double precision work array of dimension 13. c c Subprograms called c c MINPACK-2 ... dcstep c c MINPACK-1 Project. June 1983. c Argonne National Laboratory. c Jorge J. More' and David J. Thuente. c c MINPACK-2 Project. October 1993. c Argonne National Laboratory and University of Minnesota. c Brett M. Averick, Richard G. Carter, and Jorge J. More'. c c ********** double precision zero,p5,p66 parameter(zero=0.0d0,p5=0.5d0,p66=0.66d0) double precision xtrapl,xtrapu parameter(xtrapl=1.1d0,xtrapu=4.0d0) logical brackt integer stage double precision finit,ftest,fm,fx,fxm,fy,fym,ginit,gtest, + gm,gx,gxm,gy,gym,stx,sty,stmin,stmax,width,width1 c Initialization block. if (task(1:5) .eq. 'START') then c Check the input arguments for errors. if (stp .lt. stpmin) task = 'ERROR: STP .LT. STPMIN' if (stp .gt. stpmax) task = 'ERROR: STP .GT. STPMAX' if (g .ge. zero) task = 'ERROR: INITIAL G .GE. ZERO' if (ftol .lt. zero) task = 'ERROR: FTOL .LT. ZERO' if (gtol .lt. zero) task = 'ERROR: GTOL .LT. ZERO' if (xtol .lt. zero) task = 'ERROR: XTOL .LT. ZERO' if (stpmin .lt. zero) task = 'ERROR: STPMIN .LT. ZERO' if (stpmax .lt. stpmin) task = 'ERROR: STPMAX .LT. STPMIN' c Exit if there are errors on input. if (task(1:5) .eq. 'ERROR') return c Initialize local variables. brackt = .false. stage = 1 finit = f ginit = g gtest = ftol*ginit width = stpmax - stpmin width1 = width/p5 c The variables stx, fx, gx contain the values of the step, c function, and derivative at the best step. c The variables sty, fy, gy contain the value of the step, c function, and derivative at sty. c The variables stp, f, g contain the values of the step, c function, and derivative at stp. stx = zero fx = finit gx = ginit sty = zero fy = finit gy = ginit stmin = zero stmax = stp + xtrapu*stp task = 'FG' goto 1000 else c Restore local variables. if (isave(1) .eq. 1) then brackt = .true. else brackt = .false. endif stage = isave(2) ginit = dsave(1) gtest = dsave(2) gx = dsave(3) gy = dsave(4) finit = dsave(5) fx = dsave(6) fy = dsave(7) stx = dsave(8) sty = dsave(9) stmin = dsave(10) stmax = dsave(11) width = dsave(12) width1 = dsave(13) endif c If psi(stp) <= 0 and f'(stp) >= 0 for some step, then the c algorithm enters the second stage. ftest = finit + stp*gtest if (stage .eq. 1 .and. f .le. ftest .and. g .ge. zero) + stage = 2 c Test for warnings. if (brackt .and. (stp .le. stmin .or. stp .ge. stmax)) + task = 'WARNING: ROUNDING ERRORS PREVENT PROGRESS' if (brackt .and. stmax - stmin .le. xtol*stmax) + task = 'WARNING: XTOL TEST SATISFIED' if (stp .eq. stpmax .and. f .le. ftest .and. g .le. gtest) + task = 'WARNING: STP = STPMAX' if (stp .eq. stpmin .and. (f .gt. ftest .or. g .ge. gtest)) + task = 'WARNING: STP = STPMIN' cc New warning statement added to eliminate the unexpected case cc of stp=stx due to possible difference between hardware precision cc and double precision. if (stp .eq. stx) + task = 'WARNING: STP = STX' c Test for convergence. if (f .le. ftest .and. abs(g) .le. gtol*(-ginit)) + task = 'CONVERGENCE' c Test for termination. if (task(1:4) .eq. 'WARN' .or. task(1:4) .eq. 'CONV') goto 1000 c A modified function is used to predict the step during the c first stage if a lower function value has been obtained but c the decrease is not sufficient. if (stage .eq. 1 .and. f .le. fx .and. f .gt. ftest) then c Define the modified function and derivative values. fm = f - stp*gtest fxm = fx - stx*gtest fym = fy - sty*gtest gm = g - gtest gxm = gx - gtest gym = gy - gtest c Call dcstep to update stx, sty, and to compute the new step. call dcstep(stx,fxm,gxm,sty,fym,gym,stp,fm,gm, + brackt,stmin,stmax) c Reset the function and derivative values for f. fx = fxm + stx*gtest fy = fym + sty*gtest gx = gxm + gtest gy = gym + gtest else c Call dcstep to update stx, sty, and to compute the new step. call dcstep(stx,fx,gx,sty,fy,gy,stp,f,g, + brackt,stmin,stmax) endif c Decide if a bisection step is needed. if (brackt) then if (abs(sty-stx) .ge. p66*width1) stp = stx + p5*(sty - stx) width1 = width width = abs(sty-stx) endif c Set the minimum and maximum steps allowed for stp. if (brackt) then stmin = min(stx,sty) stmax = max(stx,sty) else stmin = stp + xtrapl*(stp - stx) stmax = stp + xtrapu*(stp - stx) endif c Force the step to be within the bounds stpmax and stpmin. stp = max(stp,stpmin) stp = min(stp,stpmax) c If further progress is not possible, let stp be the best c point obtained during the search. if (brackt .and. (stp .le. stmin .or. stp .ge. stmax) + .or. (brackt .and. stmax-stmin .le. xtol*stmax)) stp = stx c Obtain another function and derivative. task = 'FG' 1000 continue c Save local variables. if (brackt) then isave(1) = 1 else isave(1) = 0 endif isave(2) = stage dsave(1) = ginit dsave(2) = gtest dsave(3) = gx dsave(4) = gy dsave(5) = finit dsave(6) = fx dsave(7) = fy dsave(8) = stx dsave(9) = sty dsave(10) = stmin dsave(11) = stmax dsave(12) = width dsave(13) = width1 end c====================== The end of dcsrch ============================== subroutine dcstep(stx,fx,dx,sty,fy,dy,stp,fp,dp,brackt, + stpmin,stpmax) logical brackt double precision stx,fx,dx,sty,fy,dy,stp,fp,dp,stpmin,stpmax c ********** c c Subroutine dcstep c c This subroutine computes a safeguarded step for a search c procedure and updates an interval that contains a step that c satisfies a sufficient decrease and a curvature condition. c c The parameter stx contains the step with the least function c value. If brackt is set to .true. then a minimizer has c been bracketed in an interval with endpoints stx and sty. c The parameter stp contains the current step. c The subroutine assumes that if brackt is set to .true. then c c min(stx,sty) < stp < max(stx,sty), c c and that the derivative at stx is negative in the direction c of the step. c c The subroutine statement is c c subroutine dcstep(stx,fx,dx,sty,fy,dy,stp,fp,dp,brackt, c stpmin,stpmax) c c where c c stx is a double precision variable. c On entry stx is the best step obtained so far and is an c endpoint of the interval that contains the minimizer. c On exit stx is the updated best step. c c fx is a double precision variable. c On entry fx is the function at stx. c On exit fx is the function at stx. c c dx is a double precision variable. c On entry dx is the derivative of the function at c stx. The derivative must be negative in the direction of c the step, that is, dx and stp - stx must have opposite c signs. c On exit dx is the derivative of the function at stx. c c sty is a double precision variable. c On entry sty is the second endpoint of the interval that c contains the minimizer. c On exit sty is the updated endpoint of the interval that c contains the minimizer. c c fy is a double precision variable. c On entry fy is the function at sty. c On exit fy is the function at sty. c c dy is a double precision variable. c On entry dy is the derivative of the function at sty. c On exit dy is the derivative of the function at the exit sty. c c stp is a double precision variable. c On entry stp is the current step. If brackt is set to .true. c then on input stp must be between stx and sty. c On exit stp is a new trial step. c c fp is a double precision variable. c On entry fp is the function at stp c On exit fp is unchanged. c c dp is a double precision variable. c On entry dp is the the derivative of the function at stp. c On exit dp is unchanged. c c brackt is an logical variable. c On entry brackt specifies if a minimizer has been bracketed. c Initially brackt must be set to .false. c On exit brackt specifies if a minimizer has been bracketed. c When a minimizer is bracketed brackt is set to .true. c c stpmin is a double precision variable. c On entry stpmin is a lower bound for the step. c On exit stpmin is unchanged. c c stpmax is a double precision variable. c On entry stpmax is an upper bound for the step. c On exit stpmax is unchanged. c c MINPACK-1 Project. June 1983 c Argonne National Laboratory. c Jorge J. More' and David J. Thuente. c c MINPACK-2 Project. October 1993. c Argonne National Laboratory and University of Minnesota. c Brett M. Averick and Jorge J. More'. c c ********** double precision zero,p66,two,three parameter(zero=0.0d0,p66=0.66d0,two=2.0d0,three=3.0d0) double precision gamma,p,q,r,s,sgnd,stpc,stpf,stpq,theta sgnd = dp*(dx/abs(dx)) c First case: A higher function value. The minimum is bracketed. c If the cubic step is closer to stx than the quadratic step, the c cubic step is taken, otherwise the average of the cubic and c quadratic steps is taken. if (fp .gt. fx) then theta = three*(fx - fp)/(stp - stx) + dx + dp s = max(abs(theta),abs(dx),abs(dp)) gamma = s*sqrt((theta/s)**2 - (dx/s)*(dp/s)) if (stp .lt. stx) gamma = -gamma p = (gamma - dx) + theta q = ((gamma - dx) + gamma) + dp r = p/q stpc = stx + r*(stp - stx) stpq = stx + ((dx/((fx - fp)/(stp - stx) + dx))/two)* + (stp - stx) if (abs(stpc-stx) .lt. abs(stpq-stx)) then stpf = stpc else stpf = stpc + (stpq - stpc)/two endif brackt = .true. c Second case: A lower function value and derivatives of opposite c sign. The minimum is bracketed. If the cubic step is farther from c stp than the secant step, the cubic step is taken, otherwise the c secant step is taken. else if (sgnd .lt. zero) then theta = three*(fx - fp)/(stp - stx) + dx + dp s = max(abs(theta),abs(dx),abs(dp)) gamma = s*sqrt((theta/s)**2 - (dx/s)*(dp/s)) if (stp .gt. stx) gamma = -gamma p = (gamma - dp) + theta q = ((gamma - dp) + gamma) + dx r = p/q stpc = stp + r*(stx - stp) stpq = stp + (dp/(dp - dx))*(stx - stp) if (abs(stpc-stp) .gt. abs(stpq-stp)) then stpf = stpc else stpf = stpq endif brackt = .true. c Third case: A lower function value, derivatives of the same sign, c and the magnitude of the derivative decreases. else if (abs(dp) .lt. abs(dx)) then c The cubic step is computed only if the cubic tends to infinity c in the direction of the step or if the minimum of the cubic c is beyond stp. Otherwise the cubic step is defined to be the c secant step. theta = three*(fx - fp)/(stp - stx) + dx + dp s = max(abs(theta),abs(dx),abs(dp)) c The case gamma = 0 only arises if the cubic does not tend c to infinity in the direction of the step. gamma = s*sqrt(max(zero,(theta/s)**2-(dx/s)*(dp/s))) if (stp .gt. stx) gamma = -gamma p = (gamma - dp) + theta q = (gamma + (dx - dp)) + gamma r = p/q if (r .lt. zero .and. gamma .ne. zero) then stpc = stp + r*(stx - stp) else if (stp .gt. stx) then stpc = stpmax else stpc = stpmin endif stpq = stp + (dp/(dp - dx))*(stx - stp) if (brackt) then c A minimizer has been bracketed. If the cubic step is c closer to stp than the secant step, the cubic step is c taken, otherwise the secant step is taken. if (abs(stpc-stp) .lt. abs(stpq-stp)) then stpf = stpc else stpf = stpq endif if (stp .gt. stx) then stpf = min(stp+p66*(sty-stp),stpf) else stpf = max(stp+p66*(sty-stp),stpf) endif else c A minimizer has not been bracketed. If the cubic step is c farther from stp than the secant step, the cubic step is c taken, otherwise the secant step is taken. if (abs(stpc-stp) .gt. abs(stpq-stp)) then stpf = stpc else stpf = stpq endif stpf = min(stpmax,stpf) stpf = max(stpmin,stpf) endif c Fourth case: A lower function value, derivatives of the c same sign, and the magnitude of the derivative does not c decrease. If the minimum is not bracketed, the step is either c stpmin or stpmax, otherwise the cubic step is taken. else if (brackt) then theta = three*(fp - fy)/(sty - stp) + dy + dp s = max(abs(theta),abs(dy),abs(dp)) gamma = s*sqrt((theta/s)**2 - (dy/s)*(dp/s)) if (stp .gt. sty) gamma = -gamma p = (gamma - dp) + theta q = ((gamma - dp) + gamma) + dy r = p/q stpc = stp + r*(sty - stp) stpf = stpc else if (stp .gt. stx) then stpf = stpmax else stpf = stpmin endif endif c Update the interval which contains a minimizer. if (fp .gt. fx) then sty = stp fy = fp dy = dp else if (sgnd .lt. zero) then sty = stx fy = fx dy = dx endif stx = stp fx = fp dx = dp endif c Compute the new step. stp = stpf end c====================== The end of dcstep ============================== subroutine timer(ttime) double precision ttime c ********* c c Subroutine timer c c This subroutine is used to determine user time. In a typical c application, the user time for a code segment requires calls c to subroutine timer to determine the initial and final time. c c The subroutine statement is c c subroutine timer(ttime) c c where c c ttime is an output variable which specifies the user time. c c Argonne National Laboratory and University of Minnesota. c MINPACK-2 Project. c c Modified October 1990 by Brett M. Averick. c c ********** real temp real tarray(2) real etime c The first element of the array tarray specifies user time temp = etime(tarray) ttime = dble(tarray(1)) return end c====================== The end of timer =============================== double precision function dnrm2(n,x,incx) integer n,incx double precision x(n) c ********** c c Function dnrm2 c c Given a vector x of length n, this function calculates the c Euclidean norm of x with stride incx. c c The function statement is c c double precision function dnrm2(n,x,incx) c c where c c n is a positive integer input variable. c c x is an input array of length n. c c incx is a positive integer variable that specifies the c stride of the vector. c c Subprograms called c c FORTRAN-supplied ... abs, max, sqrt c c MINPACK-2 Project. February 1991. c Argonne National Laboratory. c Brett M. Averick. c c ********** integer i double precision scale dnrm2 = 0.0d0 scale = 0.0d0 do 10 i = 1, n, incx scale = max(scale, abs(x(i))) 10 continue if (scale .eq. 0.0d0) return do 20 i = 1, n, incx dnrm2 = dnrm2 + (x(i)/scale)**2 20 continue dnrm2 = scale*sqrt(dnrm2) return end c====================== The end of dnrm2 =============================== double precision function dpmeps() c ********** c c Subroutine dpeps c c This subroutine computes the machine precision parameter c dpmeps as the smallest floating point number such that c 1 + dpmeps differs from 1. c c This subroutine is based on the subroutine machar described in c c W. J. Cody, c MACHAR: A subroutine to dynamically determine machine parameters, c ACM Trans. Math. Soft., 14, 1988, pages 303-311. c c The subroutine statement is: c c subroutine dpeps(dpmeps) c c where c c dpmeps is a double precision variable. c On entry dpmeps need not be specified. c On exit dpmeps is the machine precision. c c MINPACK-2 Project. February 1991. c Argonne National Laboratory and University of Minnesota. c Brett M. Averick. c c ******* integer i,ibeta,irnd,it,itemp,negep double precision a,b,beta,betain,betah,temp,tempa,temp1, + zero,one,two data zero,one,two /0.0d0,1.0d0,2.0d0/ c determine ibeta, beta ala malcolm. a = one b = one 10 continue a = a + a temp = a + one temp1 = temp - a if (temp1 - one .eq. zero) go to 10 20 continue b = b + b temp = a + b itemp = int(temp - a) if (itemp .eq. 0) go to 20 ibeta = itemp beta = dble(ibeta) c determine it, irnd. it = 0 b = one 30 continue it = it + 1 b = b * beta temp = b + one temp1 = temp - b if (temp1 - one .eq. zero) go to 30 irnd = 0 betah = beta/two temp = a + betah if (temp - a .ne. zero) irnd = 1 tempa = a + beta temp = tempa + betah if ((irnd .eq. 0) .and. (temp - tempa .ne. zero)) irnd = 2 c determine dpmeps. negep = it + 3 betain = one/beta a = one do 40 i = 1, negep a = a*betain 40 continue 50 continue temp = one + a if (temp - one .ne. zero) go to 60 a = a*beta go to 50 60 continue dpmeps = a if ((ibeta .eq. 2) .or. (irnd .eq. 0)) go to 70 a = (a*(one + a))/two temp = one + a if (temp - one .ne. zero) dpmeps = a 70 return end c====================== The end of dpmeps ============================== subroutine daxpy(n,da,dx,incx,dy,incy) c c constant times a vector plus a vector. c uses unrolled loops for increments equal to one. c jack dongarra, linpack, 3/11/78. c double precision dx(*),dy(*),da integer i,incx,incy,ix,iy,m,mp1,n c if(n.le.0)return if (da .eq. 0.0d0) return if(incx.eq.1.and.incy.eq.1)go to 20 c c code for unequal increments or equal increments c not equal to 1 c ix = 1 iy = 1 if(incx.lt.0)ix = (-n+1)*incx + 1 if(incy.lt.0)iy = (-n+1)*incy + 1 do 10 i = 1,n dy(iy) = dy(iy) + da*dx(ix) ix = ix + incx iy = iy + incy 10 continue return c c code for both increments equal to 1 c c c clean-up loop c 20 m = mod(n,4) if( m .eq. 0 ) go to 40 do 30 i = 1,m dy(i) = dy(i) + da*dx(i) 30 continue if( n .lt. 4 ) return 40 mp1 = m + 1 do 50 i = mp1,n,4 dy(i) = dy(i) + da*dx(i) dy(i + 1) = dy(i + 1) + da*dx(i + 1) dy(i + 2) = dy(i + 2) + da*dx(i + 2) dy(i + 3) = dy(i + 3) + da*dx(i + 3) 50 continue return end c====================== The end of daxpy =============================== subroutine dcopy(n,dx,incx,dy,incy) c c copies a vector, x, to a vector, y. c uses unrolled loops for increments equal to one. c jack dongarra, linpack, 3/11/78. c double precision dx(*),dy(*) integer i,incx,incy,ix,iy,m,mp1,n c if(n.le.0)return if(incx.eq.1.and.incy.eq.1)go to 20 c c code for unequal increments or equal increments c not equal to 1 c ix = 1 iy = 1 if(incx.lt.0)ix = (-n+1)*incx + 1 if(incy.lt.0)iy = (-n+1)*incy + 1 do 10 i = 1,n dy(iy) = dx(ix) ix = ix + incx iy = iy + incy 10 continue return c c code for both increments equal to 1 c c c clean-up loop c 20 m = mod(n,7) if( m .eq. 0 ) go to 40 do 30 i = 1,m dy(i) = dx(i) 30 continue if( n .lt. 7 ) return 40 mp1 = m + 1 do 50 i = mp1,n,7 dy(i) = dx(i) dy(i + 1) = dx(i + 1) dy(i + 2) = dx(i + 2) dy(i + 3) = dx(i + 3) dy(i + 4) = dx(i + 4) dy(i + 5) = dx(i + 5) dy(i + 6) = dx(i + 6) 50 continue return end c====================== The end of dcopy =============================== double precision function ddot(n,dx,incx,dy,incy) c c forms the dot product of two vectors. c uses unrolled loops for increments equal to one. c jack dongarra, linpack, 3/11/78. c double precision dx(*),dy(*),dtemp integer i,incx,incy,ix,iy,m,mp1,n c ddot = 0.0d0 dtemp = 0.0d0 if(n.le.0)return if(incx.eq.1.and.incy.eq.1)go to 20 c c code for unequal increments or equal increments c not equal to 1 c ix = 1 iy = 1 if(incx.lt.0)ix = (-n+1)*incx + 1 if(incy.lt.0)iy = (-n+1)*incy + 1 do 10 i = 1,n dtemp = dtemp + dx(ix)*dy(iy) ix = ix + incx iy = iy + incy 10 continue ddot = dtemp return c c code for both increments equal to 1 c c c clean-up loop c 20 m = mod(n,5) if( m .eq. 0 ) go to 40 do 30 i = 1,m dtemp = dtemp + dx(i)*dy(i) 30 continue if( n .lt. 5 ) go to 60 40 mp1 = m + 1 do 50 i = mp1,n,5 dtemp = dtemp + dx(i)*dy(i) + dx(i + 1)*dy(i + 1) + * dx(i + 2)*dy(i + 2) + dx(i + 3)*dy(i + 3) + dx(i + 4)*dy(i + 4) 50 continue 60 ddot = dtemp return end c====================== The end of ddot ================================ subroutine dpofa(a,lda,n,info) integer lda,n,info double precision a(lda,*) c c dpofa factors a double precision symmetric positive definite c matrix. c c dpofa is usually called by dpoco, but it can be called c directly with a saving in time if rcond is not needed. c (time for dpoco) = (1 + 18/n)*(time for dpofa) . c c on entry c c a double precision(lda, n) c the symmetric matrix to be factored. only the c diagonal and upper triangle are used. c c lda integer c the leading dimension of the array a . c c n integer c the order of the matrix a . c c on return c c a an upper triangular matrix r so that a = trans(r)*r c where trans(r) is the transpose. c the strict lower triangle is unaltered. c if info .ne. 0 , the factorization is not complete. c c info integer c = 0 for normal return. c = k signals an error condition. the leading minor c of order k is not positive definite. c c linpack. this version dated 08/14/78 . c cleve moler, university of new mexico, argonne national lab. c c subroutines and functions c c blas ddot c fortran sqrt c c internal variables c double precision ddot,t double precision s integer j,jm1,k c begin block with ...exits to 40 c c do 30 j = 1, n info = j s = 0.0d0 jm1 = j - 1 if (jm1 .lt. 1) go to 20 do 10 k = 1, jm1 t = a(k,j) - ddot(k-1,a(1,k),1,a(1,j),1) t = t/a(k,k) a(k,j) = t s = s + t*t 10 continue 20 continue s = a(j,j) - s c ......exit if (s .le. 0.0d0) go to 40 a(j,j) = sqrt(s) 30 continue info = 0 40 continue return end c====================== The end of dpofa =============================== subroutine dscal(n,da,dx,incx) c c scales a vector by a constant. c uses unrolled loops for increment equal to one. c jack dongarra, linpack, 3/11/78. c modified 3/93 to return if incx .le. 0. c double precision da,dx(*) integer i,incx,m,mp1,n,nincx c if( n.le.0 .or. incx.le.0 )return if(incx.eq.1)go to 20 c c code for increment not equal to 1 c nincx = n*incx do 10 i = 1,nincx,incx dx(i) = da*dx(i) 10 continue return c c code for increment equal to 1 c c c clean-up loop c 20 m = mod(n,5) if( m .eq. 0 ) go to 40 do 30 i = 1,m dx(i) = da*dx(i) 30 continue if( n .lt. 5 ) return 40 mp1 = m + 1 do 50 i = mp1,n,5 dx(i) = da*dx(i) dx(i + 1) = da*dx(i + 1) dx(i + 2) = da*dx(i + 2) dx(i + 3) = da*dx(i + 3) dx(i + 4) = da*dx(i + 4) 50 continue return end c====================== The end of dscal =============================== subroutine dtrsl(t,ldt,n,b,job,info) integer ldt,n,job,info double precision t(ldt,*),b(*) c c c dtrsl solves systems of the form c c t * x = b c or c trans(t) * x = b c c where t is a triangular matrix of order n. here trans(t) c denotes the transpose of the matrix t. c c on entry c c t double precision(ldt,n) c t contains the matrix of the system. the zero c elements of the matrix are not referenced, and c the corresponding elements of the array can be c used to store other information. c c ldt integer c ldt is the leading dimension of the array t. c c n integer c n is the order of the system. c c b double precision(n). c b contains the right hand side of the system. c c job integer c job specifies what kind of system is to be solved. c if job is c c 00 solve t*x=b, t lower triangular, c 01 solve t*x=b, t upper triangular, c 10 solve trans(t)*x=b, t lower triangular, c 11 solve trans(t)*x=b, t upper triangular. c c on return c c b b contains the solution, if info .eq. 0. c otherwise b is unaltered. c c info integer c info contains zero if the system is nonsingular. c otherwise info contains the index of c the first zero diagonal element of t. c c linpack. this version dated 08/14/78 . c g. w. stewart, university of maryland, argonne national lab. c c subroutines and functions c c blas daxpy,ddot c fortran mod c c internal variables c double precision ddot,temp integer case,j,jj c c begin block permitting ...exits to 150 c c check for zero diagonal elements. c do 10 info = 1, n c ......exit if (t(info,info) .eq. 0.0d0) go to 150 10 continue info = 0 c c determine the task and go to it. c case = 1 if (mod(job,10) .ne. 0) case = 2 if (mod(job,100)/10 .ne. 0) case = case + 2 go to (20,50,80,110), case c c solve t*x=b for t lower triangular c 20 continue b(1) = b(1)/t(1,1) if (n .lt. 2) go to 40 do 30 j = 2, n temp = -b(j-1) call daxpy(n-j+1,temp,t(j,j-1),1,b(j),1) b(j) = b(j)/t(j,j) 30 continue 40 continue go to 140 c c solve t*x=b for t upper triangular. c 50 continue b(n) = b(n)/t(n,n) if (n .lt. 2) go to 70 do 60 jj = 2, n j = n - jj + 1 temp = -b(j+1) call daxpy(j,temp,t(1,j+1),1,b(1),1) b(j) = b(j)/t(j,j) 60 continue 70 continue go to 140 c c solve trans(t)*x=b for t lower triangular. c 80 continue b(n) = b(n)/t(n,n) if (n .lt. 2) go to 100 do 90 jj = 2, n j = n - jj + 1 b(j) = b(j) - ddot(jj-1,t(j+1,j),1,b(j+1),1) b(j) = b(j)/t(j,j) 90 continue 100 continue go to 140 c c solve trans(t)*x=b for t upper triangular. c 110 continue b(1) = b(1)/t(1,1) if (n .lt. 2) go to 130 do 120 j = 2, n b(j) = b(j) - ddot(j-1,t(1,j),1,b(1),1) b(j) = b(j)/t(j,j) 120 continue 130 continue 140 continue 150 continue return end c====================== The end of dtrsl =============================== gpml-matlab-v3.1-2010-09-27/util/lbfgsb/lbfgsb.cpp000644 000765 000024 00000007734 11422240325 021245 0ustar00carlstaff000000 000000 #include "mex.h" #include "matrix.h" #include "matlabexception.h" #include "matlabscalar.h" #include "matlabstring.h" #include "matlabmatrix.h" #include "arrayofmatrices.h" #include "program.h" #include "matlabprogram.h" #include #include extern void _main(); // Constants. // ----------------------------------------------------------------- const int minNumInputArgs = 5; // Function definitions. // ----------------------------------------------------------------- void mexFunction (int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) try { // Check to see if we have the correct number of input and output // arguments. if (nrhs < minNumInputArgs) throw MatlabException("Incorrect number of input arguments"); // Get the starting point for the variables. This is specified in // the first input argument. The variables must be either a single // matrix or a cell array of matrices. int k = 0; // The index of the current input argument. ArrayOfMatrices x0(prhs[k++]); // Create the output, which stores the solution obtained from // running IPOPT. There should be as many output arguments as cell // entries in X. if (nlhs != x0.length()) throw MatlabException("Incorrect number of output arguments"); ArrayOfMatrices x(plhs,x0); // Load the lower and upper bounds on the variables as // ArrayOfMatrices objects. They should have the same structure as // the ArrayOfMatrices object "x". ArrayOfMatrices lb(prhs[k++]); ArrayOfMatrices ub(prhs[k++]); // Check to make sure the bounds make sense. if (lb != x || ub != x) throw MatlabException("Input arguments LB and UB must have the same \ structure as X"); // Get the Matlab callback functions. MatlabString objFunc(prhs[k++]); MatlabString gradFunc(prhs[k++]); // Get the auxiliary data. const mxArray* auxData; const mxArray* ptr = prhs[k++]; if (nrhs > 5) { if (mxIsEmpty(ptr)) auxData = 0; else auxData = ptr; } else auxData = 0; // Get the intermediate callback function. MatlabString* iterFunc; ptr = prhs[k++]; if (nrhs > 6) { if (mxIsEmpty(ptr)) iterFunc = 0; else iterFunc = new MatlabString(ptr); } else iterFunc = 0; // Set the options for the L-BFGS algorithm to their defaults. int maxiter = defaultmaxiter; int m = defaultm; double factr = defaultfactr; double pgtol = defaultpgtol; // Process the remaining input arguments, which set options for // the IPOPT algorithm. while (k < nrhs) { // Get the option label from the Matlab input argument. MatlabString optionLabel(prhs[k++]); if (k < nrhs) { // Get the option value from the Matlab input argument. MatlabScalar optionValue(prhs[k++]); double value = optionValue; if (!strcmp(optionLabel,"maxiter")) maxiter = (int) value; else if (!strcmp(optionLabel,"m")) m = (int) value; else if (!strcmp(optionLabel,"factr")) factr = value / mxGetEps(); else if (!strcmp(optionLabel,"pgtol")) pgtol = value; else { if (iterFunc) delete iterFunc; throw MatlabException("Nonexistent option"); } } } // Create a new instance of the optimization problem. x = x0; MatlabProgram program(x,lb,ub,&objFunc,&gradFunc,iterFunc, (mxArray*) auxData,m,maxiter,factr,pgtol); // Run the L-BFGS-B solver. SolverExitStatus exitStatus = program.runSolver(); if (exitStatus == abnormalTermination) { if (iterFunc) delete iterFunc; throw MatlabException("Solver unable to satisfy convergence \ criteria due to abnormal termination"); } else if (exitStatus == errorOnInput) { if (iterFunc) delete iterFunc; throw MatlabException("Invalid inputs to L-BFGS routine"); } // Free the dynamically allocated memory. if (iterFunc) delete iterFunc; } catch (std::exception& error) { mexErrMsgTxt(error.what()); } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabstring.h000644 000765 000024 00000002202 11422240325 022123 0ustar00carlstaff000000 000000 #ifndef INCLUDE_MATLABSTRING #define INCLUDE_MATLABSTRING #include "mex.h" #include // Function declarations. // ----------------------------------------------------------------- // Copy a C-style string (i.e. a null-terminated character array). char* copystring (const char* source); // Class MatlabString. // ----------------------------------------------------------------- // This class encapsulates read-only access to a MATLAB character // array. class MatlabString { public: // The constructor accepts as input a pointer to a Matlab array, // which must be a valid string (array of type CHAR). explicit MatlabString (const mxArray* ptr); // The copy constructor makes a full copy of the source string. MatlabString (const MatlabString& source); // The destructor. ~MatlabString(); // Conversion operator for null-terminated string. operator const char* () const { return s; }; protected: char* s; // The null-terminated string. // The copy assignment operator is not proper, thus remains // protected. MatlabString& operator= (const MatlabString& source) { return *this; }; }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/README.html000644 000765 000024 00000043521 11422240325 021117 0ustar00carlstaff000000 000000 MATLAB interface for L-BFGS-B

A MATLAB interface for L-BFGS-B

by Peter Carbonetto
Dept. of Computer Science
University of British Columbia

L-BFGS-B is a collection of Fortran 77 routines for solving large-scale nonlinear optimization problems with bound constraints on the variables. One of the key features of the nonlinear solver is that knowledge of the Hessian is not required; the solver computes search directions by keeping track of a quadratic model of the objective function with a limited-memory BFGS (Broyden-Fletcher-Goldfarb-Shanno) approximation to the Hessian.1 The algorithm was developed by Ciyou Zhu, Richard Byrd and Jorge Nocedal. For more information, go to the original distribution site for the L-BFGS-B software package.

I've designed an interface to the L-BFGS-B solver so that it can be called like any other function in MATLAB.2 See the text below for more information on installing and calling it in MATLAB. Along the way, I've also developed a C++ class that encapsulates all the "messy" details in executing the L-BFGS-B code. See below for instructions on how to use this class for your own C++ code.

If you have any questions, praise, or comments, or would like to report a bug, do not hesitate to contact the author. I've tested this software in MATLAB version 7.2 on both the Mac OS X and Linux operating systems.

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 2.5 Canada License. Please cite the source as Peter Carbonetto, Department of Computer Science, University of British Columbia.

Installation

These installation instructions assume you have a UNIX-based operating system, such as Mac OS X or Linux. It is surely possible to install this software on Windows, I'm just not sure how. These instructions also assume you have GNU make installed.

Download the source. First, download and unpack the tar archive. It contains some MATLAB files (ending in .m), some C++ source and header files (ending in .h and .cpp), a single Fortran 77 source file solver.f containing the L-BFGS-B routines, and a Makefile. I've included a version of the Fortran routines that is more recent than what is available for download at the distribution site at Northwestern University.

What we will do is create a MEX file, which is basically a file that contains a routine that can be called from MATLAB as if it were a built-it function. To learn about MEX files, I refer you to this document at the MathWorks website.

Install the C++ and Fortran 77 compilers. In order to build the MEX file, you will need both a C++ compiler and a Fortran 77 compiler. Unfortunately, you can't just use any compiler. You have to use the precise one supported by MATLAB. For instance, if you are running Mac OS X 7.2 on a Linux operating system, you will need to install the GNU compiler collection (GCC) version 3.4.4. Even if you already have a compiler installed on Linux, it may be the wrong version, and if you use the wrong version things could go horribly wrong! It is important that you use the correct version of the compiler, otherwise you will encounter linking errors. Refer to this document to find out which compiler is supported by your version of MATLAB.

Configure MATLAB. Next, you need to set up and configure MATLAB to build MEX Files. This is explained quite nicely in this MathWorks support document.

Modify the Makefile. You are almost ready to build the MEX file. But before you do so, you need to edit the Makefile to coincide with your system setup. At the top of the file are several variables you may need to modify. The variable MEX is the executable called to build the MEX file. The variables CXX and F77 must be the names of your C++ and Fortran 77 compilers. MEXSUFFIX is the MEX file extension specific to your platform. (For instance, the extension for Linux is mexglx.) The variable MATLAB_HOME must be the base directory of your MATLAB installation. Finally, CFLAGS and FFLAGS are options passed to the C++ and Fortran compilers, respectively. Often these flags coincide with your MEX options file (see here for more information on the options file). But often the settings in the MEX option file are incorrect, and so the options must be set manually. MATLAB requires some special compilation flags for various reasons, one being that it requires position-independent code. These instructions are vague (I apologize), and this step may require a bit of trial and error until before you get it right.

On my laptop running Mac OS X 10.3.9 with MATLAB 7.2, my settings for the Makefile are

  MEX         = mex
  MEXSUFFIX   = mexmac
  MATLAB_HOME = /Applications/MATLAB72
  CXX         = g++
  F77         = g77 
  CFLAGS      = -O3 -fPIC -fno-common -fexceptions \
                -no-cpp-precomp 
  FFLAGS      = -O3 -x f77-cpp-input -fPIC -fno-common

On my Linux machine, I set the variables in the Makefile like so:

  MEX         = mex
  MEXSUFFIX   = mexglx
  MATLAB_HOME = /cs/local/generic/lib/pkg/matlab-7.2
  CXX         = g++-3.4.5
  F77         = g77-3.4.5
  CFLAGS      = -O3 -fPIC -pthread 
  FFLAGS      = -O3 -fPIC -fexceptions

It may be helpful to look at the GCC documentation in order to understand what these various compiler flags mean.

Build the MEX file. If you are in the directory containing all the source files, typing make in the command prompt will first compile the Fortran and C++ source files into object code (.o files). After that, the make program calls the MEX script, which in turn links all the object files together into a single MEX file. If you didn't get any errors, then you are ready to try out the bound-constrained solver in MATLAB. Note that even if you didn't get any errors, there's still a possibility that you didn't link the MEX file properly, in which case executing the MEX file will cause MATLAB to crash. But there's only one way to find out: the hard way.

A brief tutorial

I've written a short script examplehs038.m which demonstrates how to call lbfgsb for solving a small optimization problem, the Hock & Schittkowski test problem no. 38.3 The optimization problem has four variables. They are bounded from below by -10 and bounded from above by +10. We've set the starting point to (-3, -1, -3, -1). We've written two MATLAB functions for computing the objective function and the gradient of the objective function at a given point. These functions can be found in the files computeObjectiveHS038.m and computeGradientHS038.m. If you run the script, the solver should very quickly progress toward the point (1, 1, 1, 1). The objective is 0 at this point.

The basic MATLAB function call is

  lbfgsb(x0,lb,ub,objfunc,gradfunc)

The first input argument x0 declares the starting point for the solver. The second and third input arguments define the lower and upper bounds on the variables, respectively. If an entry of lb is set to -Inf, then this is equivalent to no lower bound for that entry (the same applies to upper bounds, except that it must be +Inf instead). The next two input arguments must be the names of MATLAB routines (M-files). The first routine must calculate the objective function at the current point. The output must be a scalar representing the objective evaluated at the current point. The second function must compute the gradient of the objective at the current point. The input is the same as objfunc, but it must return as many outputs as there are inputs. For a complete guide on using the MATLAB interface, type help lbfgsb in the MATLAB prompt.

The script exampleldaimages.m is a more complicated example. It uses the L-BFGS-B solver to compute some statistics of a posterior distribution for a text analysis model called latent Dirichlet allication (LDA). The model is used to estimate the topics for a collection of documents. The optimization problem is to compute an approximation to the posterior, since it is not possible to compute the posterior exactly. In this case, the optimization problem corresponds to minimizing the distance between the true posterior distribution and an approximate distribution. For more information, see the paper of Blei, Ng and Jordan (2003).4 In order to run this example, you will need to install the lightspeed toolbox for MATLAB developed by Tom Minka at Microsoft Research.

The script starts by generating a bunch of canonical images that represent topics; if a pixel is white then a word at that position is more likely to appear in that topic. Note that the order of the pixels in the image is unimportant, since LDA is a "bag of words" model. In another figure we show a small sample of the data generated from the topics. Each image represents a document, and the pixels in the image depict the word proportions in that document.5

Next, the script runs the L-BFGS-B solver to find a local minimum to the variational objective. Buried in the M-file mflda.m is the call to the solver: 

  [xi gamma Phi] = lbfgsb({xi gamma Phi},...
      {repmat(lb,W,K)  repmat(lb,K,D)  repmat(lb,K,M)},...
      {repmat(inf,W,K) repmat(inf,K,D) repmat(inf,K,M)},...
      'computeObjectiveMFLDA','computeGradientMFLDA',...
      {nu eta L w lnZconst},'callbackMFLDA',...
      'm',m_lbfgs,'factr',1e-12,'pgtol',...
      tolerance,'maxiter',maxiter);

This example will give us the opportunity to demonstrate some of the more complicated aspects of the our MATLAB interface. The first input to lbfgsb sets the initial iterate of the optimization algorithm. Notice that we've passed a cell array with entries that are matrices. Indeed, it is possible to pass either a matrix or a cell array. Whatever structure is passed, the other input arguments (and outputs) must also abide by this structure. The lower and upper bounds on variables are also cell arrays. All the upper bounds are set to infinity, which means that the variables are only bounded from below.

It is instructive to examine the callback routines. They look like: 

  f = computeObjectiveMFLDA(xi,gamma,Phi,auxdata)

  [dxi, dgamma, dPhi] = ...
      computeGradientMFLDA(xi,gamma,Phi,auxdata)

Notice that each entry to the cell array is its own input argument to the callback routines. The same applies for the output arguments of the gradient callback function.

The sixth input argument is a cell array that contains auxiliary data. This data will be passed to the MATLAB functions that evaluate the objective and gradient. The seventh input argument specifies a callback routine that is called exactly once for every iteration. This callback routine is most useful for examining the progress of the solver. The remaining input arguments are label/value pairs that override some of the default settings of L-BFGS-B.

After converging to a solution, we display topic samples drawn from the variational approximation. A good solution should come close to recovering the true topics from which the data was generated (although they might not be in the same order).

Don't be surprised if it takes many iterations before the optimization algorithm converges on a local minimum. (In repeated trials, I found it as many as two thousand iterations to converge to a stationary point of the objective.) This particular problem demonstrates the limitations of limited-memory quasi-Newton approximations to the Hessian; the low storage requirements can come at the cost of slow convergence to the solution.

The C++ interface

One of the nice byproducts of writing a MATLAB interface for L-BFGS-B is that I've ended up with a neat little C++ class that encapsulates the nuts and bolts of executing the solver. A brief description of the Program class can be found in the header file program.h.

Program is an abstract class, which means that its impossible to instantiate a Program object. This is because the class has member functions that aren't yet implemented. These are called pure virtual functions. In order to use the Program class, one needs to define another class that inherits the Program class and that implements the pure virtual functions. In fact, the class MatlabProgram (in matlabprogram.h) is an example of such a class.

The new child class must declare and implement these two functions:

  virtual double computeObjective (int n, double* x);
  virtual void computeGradient (int n, double* x, double* g);

See the header file for a detailed description of these functions. The only remaining detail is calling the Program constructor. After that, it is just a matter of declaring a new object of type MyProgram then calling the function runSolver.

Contents of the tar archive

solver.f
Fortran 77 source for the L-BFGS-B solver routines.

program.h
program.cpp
Header and source file for the C++ interface.

array.h
arrayofmatrices.h
matlabexception.h
matlabmatrix.h
matlabprogram.h
matlabscalar.h
matlabstring.h
arrayofmatrices.cpp
matlabexception.cpp
matlabmatrix.cpp
matlabprogram.cpp
matlabscalar.cpp
matlabstring.cpp
lbfgsb.cpp
Header and source files for the MATLAB interface.

lbfgsb.m
Usage instructions for the MEX file.

examplehs038.m
MATLAB script for solving the H&S test example no. 38.

computeObjectiveHS038.m
computeGradientHS038.m
genericcallback.m
MATLAB functions used by the script examplehs038.m.

exampleldaimages.m
MATLAB script that generates synthetic documents and topics, then computes a mean field variational approximation to the posterior distribution of the latent Dirichlet allocation model, then displays the result.

mflda.m
This function computes a mean field variational approximation to the posterior distribution of the latent Dirichlet allocation model by minimizing the distance between the variational distribution and the true distribution. It uses the L-BFGS-B to minimize the objective function subject to the bound constraints. The objective function also acts as a lower bound on the logarithm of the denominator that appears from the application of Bayes' rule.4

callbackMFLDA.m
computeObjectiveMFLDA.m
computeGradientMFLDA.m
computelnZ.m
computelnZconst.m
computem.m
dirichletrnd.m
gammarnd.m
createsyntheticdata.m
mfldainit.m
Some functions used by exampleldaimages.m and mflda.m.

Footnotes

1 Ciyou Zhu, Richard H. Byrd, Peihuang Lu and Jorge Nocedal (1997). Algorithm 778: L-BFGS-B: Fortran subroutines for large-scale bound-constrained optimization. ACM Transactions on Mathematical Software, Vol. 23, No. 4., pp. 550-560.

1 Another MATLAB interface for the L-BFGS-B routines has been developed by Liam Stewart at the University of Toronto.

3 Willi Hock and Klaus Schittkowski (1981). Test Examples for Nonlinear Programming Codes. Lecture Notes in Economics and Mathematical Systems, Vol. 187, Springer-Verlag.

4 David M. Blei, Andrew Y. Ng, Michael I. Jordan (2003). Latent Dirichlet allocation. Journal of Machine Learning Research, Vol. 3, pp. 993-1022.

5 Tom L. Griffiths and Mark Steyvers (2004). Finding scientific topics. Proceedings of the National Academy of Sciences, Vol. 101, pp. 5228-5235.


March 3, 2007
gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabprogram.h000644 000765 000024 00000005501 11422240325 022271 0ustar00carlstaff000000 000000 #ifndef INCLUDE_MATLABPROGRAM #define INCLUDE_MATLABPROGRAM #include "mex.h" #include "program.h" #include "matlabscalar.h" #include "matlabstring.h" #include "arrayofmatrices.h" // Class MatlabProgram. // ----------------------------------------------------------------- // This is an implementation of the abstract class Program. class MatlabProgram: public Program { public: // On input, "variables" should contain the initial point for the // optimization routine. If no iterative callback function is // specified, "iterFunc" may be set to 0. Also, if no auxiliary data // is needed, it may also be set to 0. MatlabProgram (ArrayOfMatrices& variables, const ArrayOfMatrices& lowerbounds, const ArrayOfMatrices& upperbounds, const MatlabString* objFunc, const MatlabString* gradFunc, const MatlabString* iterFunc, mxArray* auxData, int m = defaultm, int maxiter = defaultmaxiter, double factr = defaultfactr, double pgtol = defaultpgtol); // The destructor. virtual ~MatlabProgram(); // These provide definitions for the pure virtual functions of the // abstract parent class. virtual double computeObjective (int n, double* x); virtual void computeGradient (int n, double* x, double* g); virtual void iterCallback (int t, double* x, double f); // Run the solver. Upon completion, the solution is stored in // "variables". SolverExitStatus runSolver(); protected: ArrayOfMatrices& variables; // Storage for the initial value and // solution. const MatlabString* objFunc; // The objective callback function. const MatlabString* gradFunc; // The gradient callback function. const MatlabString* iterFunc; // The iterative callback function. ArrayOfMatrices* varMatlab; // Inputs to the Matlab callback mxArray** varInputs; // functions representing the // current values of the variables. MatlabScalar* fMatlab; // Input to the Matlab callback // functions representing the // current value of the objective. MatlabScalar* tMatlab; // Input to the Matlab callback // functions representing the // current iteration. int numInputsObjFunc; // The number of inputs passed to the // objective callback function. int numInputsGradFunc; // The number of inputs passed to the // gradient callback function. int numInputsIterFunc; // The number of inputs passed to the // iterative callback function. mxArray** inputsObjFunc; // The inputs passed to the objective // callback function. mxArray** inputsGradFunc; // The inputs passed to the gradient // callback function. mxArray** inputsIterFunc; // The inputs passed to the iterative // callback function. }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/arrayofmatrices.h000644 000765 000024 00000004144 11422240325 022636 0ustar00carlstaff000000 000000 #ifndef INCLUDE_ARRAYOFMATRICES #define INCLUDE_ARRAYOFMATRICES #include "array.h" #include "matlabmatrix.h" #include "mex.h" // Class ArrayOfMatrices. // ----------------------------------------------------------------- class ArrayOfMatrices : public Array { public: // This version of the constructor behaves just like its parent. explicit ArrayOfMatrices (int length) : Array(length) { }; // This constructor creates an array of matrices from a Matlab // array. It accepts either a matrix in double precision, or a cell // array with entries that are matrices. explicit ArrayOfMatrices (const mxArray* ptr); // This constructor creates an array of matrices from a collection // of Matlab arrays. The Matlab arrays must be matrices. ArrayOfMatrices (const mxArray* ptrs[], int numptrs); // This constructor creates an array of matrices and the // associated Matlab structures. The Matlab structures are // matrices. The second input argument acts as a template for the // creation of the matrices, but the data from "model" is not // actually copied into the new ArrayOfMatrices object. It is up // to the user to make sure that the array of mxArray pointers has // enough room for the pointers to the Matlab arrays. ArrayOfMatrices (mxArray* ptrs[], const ArrayOfMatrices& model); // This constructor creates an array of matrices using the second // input argument as a model. The input argument "data" contains // the element data. Note that the information is NOT copied from // the model! ArrayOfMatrices (double* data, const ArrayOfMatrices& model); // The copy constructor makes a shallow copy of the source object. ArrayOfMatrices (const ArrayOfMatrices& source); // The destructor. ~ArrayOfMatrices(); // Copy assignment operator that observes the same behaviour as // the Array copy assignment operator. ArrayOfMatrices& operator= (const ArrayOfMatrices& source); // Return the total number of elements in all the matrices. int numelems() const; protected: static int getnummatlabmatrices (const mxArray* ptr); }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/array.h000644 000765 000024 00000010714 11422240325 020561 0ustar00carlstaff000000 000000 #ifndef INCLUDE_ARRAY #define INCLUDE_ARRAY #include "matlabexception.h" #include // Function definitions. // ----------------------------------------------------------------- template void copymemory (const Type* source, Type* dest, int length) { memcpy(dest,source,sizeof(Type)*length); } // Class Array // ----------------------------------------------------------------- // An Array object stores a collection of ordered elements. The key // aspect of Array objects is that they do not necessarily achieve // encapsulation; they do not necessarily retain independent // ownership of their data. That is, the data could be modified // externally by another agent. This behaviour is determined by the // data member "owner". If "owner" is false, the array does not take // care of allocation and deallocation of the data in memory. // // Copy assignment behaves different than usual---it copies the // data, but requires that the destination already have the proper // resources allocated in memory. // // The copy constructor performs a shallow copy. template class Array { public: // This constructor allocates memory for an array of elements of // the specificed length. explicit Array (int length); // This constructor is set to point to an already existing // array. As such, it does not gain ownership of the elements. Array (Type* data, int length); // The copy constructor performs a shallow copy, so that both // arrays share the same copy of the data. Array (const Array& source); // The destructor. ~Array(); // Set all the elements of the array to the specified value. void setvalue (const Type& value); // Copy the elements to from the source array. Both the source and // destination array must have the same length. void inject (const Type* source); void inject (const Array& source); Array& operator= (const Type* source); Array& operator= (const Array& source); // Get the number of elements in the array. int length() const { return n; }; // Elements of an array A can be accessed the subscript operator; // e.g. A[i]. For the sake of efficiency, no checks are made to // ensure that the subscripting is legal. It is up to the user to // ensure that i is non-negative and less than A.length(). Note // that subscripts, as per C++ convention, start at 0. Type& operator[] (int i); const Type& operator[] (int i) const; // Returns true if the two arrays have the same dimensions // (i.e. the same lengths). bool operator== (const Array& a) const { return n == a.n; }; bool operator!= (const Array& a) const { return !(*this == a); }; protected: Type* elems; int n; // The number of elements in the array. bool owner; // Whether the object has ownership of the data. }; // Function definitions for class Array. // ----------------------------------------------------------------- template Array::Array (int length) { n = length; owner = true; elems = 0; elems = new Type[length]; } template Array::Array (Type* data, int length) { n = length; owner = false; elems = data; } template Array::Array (const Array& source) { n = source.n; owner = false; elems = source.elems; } template Array::~Array () { if (owner && elems) delete[] elems; } template void Array::setvalue (const Type& value) { for (int i = 0; i < n; i++) elems[i] = value; } template void Array::inject (const Type* source) { // Check for self-assignment. If there is self-assignment, there // is no need to copy because the objects share the same data! if (elems != source) copymemory(source,elems,n); } template void Array::inject (const Array& source) { if (n != source.n) throw MatlabException("Unable to perform copy; arrays have \ different lengths"); inject(source.elems); } template Array& Array::operator= (const Type* source) { inject(source); return *this; } template Array& Array::operator= (const Array& source) { inject(source); return *this; } template Type& Array::operator[] (int i) { return elems[i]; } template const Type& Array::operator[] (int i) const { return elems[i]; } #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabexception.cpp000644 000765 000024 00000001000 11422240325 023141 0ustar00carlstaff000000 000000 #include "matlabexception.h" // Function definitions for class MatlabException. // ----------------------------------------------------------------- MatlabException::MatlabException (const char* message) throw() : exception() { this->message = message; } MatlabException::MatlabException (const MatlabException& source) throw() : exception() { message = source.message; } MatlabException& MatlabException::operator= (const MatlabException& source) { message = source.message; return *this; } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/arrayofmatrices.cpp000644 000765 000024 00000004655 11422240325 023200 0ustar00carlstaff000000 000000 #include "arrayofmatrices.h" #include "matrix.h" // Function definitions for class ArrayOfMatrices. // ----------------------------------------------------------------- ArrayOfMatrices::ArrayOfMatrices (const mxArray* ptr) : Array(getnummatlabmatrices(ptr)) { setvalue(0); if (mxIsCell(ptr)) // Fill in the entries of the array with the entries of the Matlab // cell array. for (int i = 0; i < n; i++) elems[i] = new Matrix(mxGetCell(ptr,i)); else // Fill in the single entry of the array with the Matlab matrix. elems[0] = new Matrix(ptr); } ArrayOfMatrices::ArrayOfMatrices (const mxArray* ptrs[], int numptrs) : Array(numptrs) { setvalue(0); for (int i = 0; i < numptrs; i++) elems[i] = new Matrix(ptrs[i]); } ArrayOfMatrices::ArrayOfMatrices (mxArray* ptrs[], const ArrayOfMatrices& source) : Array(source.length()) { setvalue(0); // Create the new Matlab matrices and copy the data from the // source. for (int i = 0; i < n; i++) { int h = source[i]->height(); int w = source[i]->width(); elems[i] = new Matrix(ptrs[i],h,w); *elems[i] = *source[i]; } } ArrayOfMatrices::ArrayOfMatrices (const ArrayOfMatrices& source) : Array(source) { } ArrayOfMatrices::ArrayOfMatrices (double* data, const ArrayOfMatrices& model) : Array(model.length()) { setvalue(0); // Repeat for each matrix. for (int i = 0; i < n; i++) { const Matrix* A = model[i]; elems[i] = new Matrix(data,A->height(),A->width()); data += A->length(); } } ArrayOfMatrices::~ArrayOfMatrices() { for (int i = 0; i < n; i++) if (elems[i]) delete elems[i]; } int ArrayOfMatrices::numelems() const { int m = 0; // The return value. // Repeat for each matrix. for (int i = 0; i < n; i++) m += elems[i]->length(); return m; } ArrayOfMatrices& ArrayOfMatrices::operator= (const ArrayOfMatrices& source) { // Copy each matrix. for (int i = 0; i < n; i++) *elems[i] = *source.elems[i]; return *this; } int ArrayOfMatrices::getnummatlabmatrices (const mxArray* ptr) { int n; // The return value. if (mxIsCell(ptr)) n = mxGetNumberOfElements(ptr); else if (mxGetNumberOfDimensions(ptr) == 2 && mxIsDouble(ptr)) n = 1; else throw MatlabException("The Matlab array must either be a cell \ array or a double precision matrix"); return n; } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/program.h000644 000765 000024 00000011647 11422240325 021120 0ustar00carlstaff000000 000000 #ifndef INCLUDE_PROGRAM #define INCLUDE_PROGRAM // Type definitions. // ----------------------------------------------------------------- // This defines the possible results of running the L-BFGS-B solver. enum SolverExitStatus { success, // The algorithm has converged to a stationary // point or has reached the maximum number of // iterations. abnormalTermination, // The algorithm has terminated abnormally and // was unable to satisfy the convergence // criteria. errorOnInput // The routine has detected an error in the // input parameters. }; // Constants. // ----------------------------------------------------------------- const int defaultm = 5; const int defaultmaxiter = 100; const double defaultfactr = 1e7; const double defaultpgtol = 1e-5; const int defaultprintlevel = -1; // Function declarations. // ----------------------------------------------------------------- // This is the L-BFGS-B routine implemented in Fortran 77. extern "C" void setulb_ (int* n, int* m, double x[], double l[], double u[], int nbd[], double* f, double g[], double* factr, double* pgtol, double wa[], int iwa[], char task[], int* iprint, char csave[], bool lsave[], int isave[], double dsave[]); // Class Program. // ----------------------------------------------------------------- // This class encapsulates execution of the L-BFGS-B routine for // solving a nonlinear optimization problem with bound constraints // using limited-memory approximations to the Hessian. // // This is an abstract class since some of the class methods have not // been implemented (they are "pure virtual" functions). In order to // use this class, one needs to define a child class and provide // definitions for the pure virtual methods. class Program { public: // The first input argument "n" is the number of variables, and "x" // must be set to the suggested starting point for the optimization // algorithm. See the L-BFGS-B documentation for more information on // the inputs to the constructor. Program (int n, double* x, double* lb, double* ub, int* btype, int m = defaultm, int maxiter = defaultmaxiter, double factr = defaultfactr, double pgtol = defaultpgtol); // This is the same constructor as above, except that the // appropriate amount of memory is dynamically allocated for the // variables, the bounds, and the bound types. The destructor also // makes sure that the memory is properly deallocated, but not so // for the other constructor. Program (int n, int m = defaultm, int maxiter = defaultmaxiter, double factr = defaultfactr, double pgtol = defaultpgtol); // The destructor. virtual ~Program(); // An implementation of this method should return the value of the // objective at the current value of the variables "x", where "n" is // the number of variables. virtual double computeObjective (int n, double* x) = 0; // An implementation of this method should fill in the values of the // gradient "g" and the current point "x". virtual void computeGradient (int n, double* x, double* g) = 0; // The child class may optionally override this method which is // called once per iteration of the L-BFGS-B routine. The input // arguments are the current iteration "t", the current value of the // variables "x" and the current value of the objective "f". virtual void iterCallback (int t, double* x, double f) { }; // Run the solver. Upon completion, the solution is stored in "x". SolverExitStatus runSolver(); protected: // The copy constructor and copy assignment operator are kept // private so that they are not used. Program (const Program& source) { }; Program& operator= (const Program& source) { return *this; }; int n; // The number of variables. double* x; // The current point. double* lb; // The lower bounds. double* ub; // The upper bounds. int* btype; // The bound types. private: // Execute a single step the L-BFGS-B solver routine. void callLBFGS (const char* cmd = 0); // Initialize some of the structures used by the L-BFGS-B routine. void initStructures(); bool owner; // If true, then the blocks of memory pointed to by // x, lb, ub and btype must be deallocated in the // destructor. double f; // The value of the objective. double* g; // The value of the gradient. int iprint; // The print level. int maxiter; // The maximum number of iterations. double pgtol; // Convergence parameter passed to L-BFGS-B. double factr; // Convergence parameter passed to L-BFGS-B. int m; // The number of variable corrections to the // limited-memory approximation to the Hessian. // These are structures used by the L-BFGS-B routine. double* wa; int* iwa; char task[60]; char csave[60]; bool lsave[4]; int isave[44]; double dsave[29]; }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/license.txt000644 000765 000024 00000003027 11422240325 021454 0ustar00carlstaff000000 000000 Copyright (c) 2009, Peter Carbonetto All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution * Neither the name of the University of British Columbia nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabscalar.h000644 000765 000024 00000002264 11422240325 022072 0ustar00carlstaff000000 000000 #ifndef INCLUDE_MATLABSCALAR #define INCLUDE_MATLABSCALAR #include "mex.h" // Class MatlabScalar // ----------------------------------------------------------------- // The main appeal of this class is that one can create a scalar // object that accesses a MATLAB array. // // Note that the copy assignment operator is not valid for this class // because we cannot reassign a reference. class MatlabScalar { public: // This constructor accepts as input a pointer to a Matlab array // which must be a scalar in double precision. explicit MatlabScalar (const mxArray* ptr); // This constructor creates a new Matlab array which is a scalar // in double precision. MatlabScalar (mxArray*& ptr, double value); // The copy constructor. MatlabScalar (MatlabScalar& source); // The destructor. ~MatlabScalar() { }; // Access the value of the scalar. operator const double () const { return x; }; // Assign the value of the scalar. MatlabScalar& operator= (double value); protected: double& x; // The copy assignment operator is kept protected because it is // invalid. MatlabScalar& operator= (const MatlabScalar& source) { return *this; }; }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabmatrix.cpp000644 000765 000024 00000003156 11422240325 022465 0ustar00carlstaff000000 000000 #include "matlabmatrix.h" #include "matlabexception.h" // Function definitions. // ----------------------------------------------------------------- double* getMatlabMatrixDouble (const mxArray* ptr) { if (mxGetNumberOfDimensions(ptr) != 2) throw MatlabException("Matlab array must be a matrix"); if (!mxIsDouble(ptr)) throw MatlabException("Matlab array must be of type double"); return mxGetPr(ptr); } double* createMatlabMatrixDouble (mxArray*& ptr, int height, int width) { ptr = mxCreateDoubleMatrix(height,width,mxREAL); return mxGetPr(ptr); } // Function definitions for class Matrix. // ----------------------------------------------------------------- Matrix::Matrix (int height, int width) : Array(height*width) { h = height; w = width; } Matrix::Matrix (double* data, int height, int width) : Array(data,height*width) { h = height; w = width; } Matrix::Matrix (const mxArray* ptr) : Array(getMatlabMatrixDouble(ptr),mxGetNumberOfElements(ptr)) { h = mxGetM(ptr); w = mxGetN(ptr); } Matrix::Matrix (mxArray*& ptr, int height, int width) : Array(createMatlabMatrixDouble(ptr,height,width), height*width) { h = height; w = width; } Matrix::Matrix (const Matrix& source) : Array(source) { h = source.h; w = source.w; } Matrix& Matrix::operator= (const Matrix& source) { inject(source); return *this; } bool Matrix::operator== (const Matrix& X) const { return (h == X.h) && (w == X.w); } double& Matrix::entry (int r, int c) { return elems[h*c + r]; } double Matrix::entry (int r, int c) const { return elems[h*c + r]; } gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabmatrix.h000644 000765 000024 00000004562 11422240325 022134 0ustar00carlstaff000000 000000 #ifndef INCLUDE_MATLABMATRIX #define INCLUDE_MATLABMATRIX #include "array.h" #include "mex.h" // Class Matrix // --------------------------------------------------------------- // A matrix object stores its elements in column-major format, as in // Fortran and Matlab. This means that columns are stored one after // another. For example, the matrix // // 1 2 3 // 4 5 6 // // is stored in memory as // // 1 4 2 5 3 6 // // Like Array objects, Matrix objects are not necessarily // encapsulated, and they exhibit analogous behaviour. class Matrix : public Array { public: // This constructor allocates memory for matrix of the specified // heigth and width. Matrix (int height, int width); // This constructor basically follows the lead of the analagous // constructor for the Array class. Matrix (double* data, int height, int width); // This constructor retrieves a matrix from a Matlab array. This // particular constructor is only defined for Matrix. // Since Matlab handles storage, the object created by this // constructor is not encapsulated. explicit Matrix (const mxArray* ptr); // This constructor creates a new Matlab array as a side effect. // Since Matlab handles storage, the object created by this // constructor is not encapsulated. Matrix (mxArray*& ptr, int height, int width); // The copy constructor makes a shallow copy of the data. Matrix (const Matrix& source); // The destructor. ~Matrix() { }; // Copy assignment operator that observes the same behaviour as // the Array copy assignment operator. Matrix& operator= (const Matrix& source); // Get the height and width of the matrix. int height() const { return h; }; int width () const { return w; }; // Returns true if the two matrices have the same dimensions // (i.e. the same height and width). bool operator== (const Matrix& X) const; bool operator!= (const Matrix& X) const { return !(*this == X); }; // If X is an object of type Matrix, X.entry(r,c) accesses the // entry of the rth row and cth column. double entry (int r, int c) const; double& entry (int r, int c); double operator() (int r, int c) const { return entry(r,c); }; double& operator() (int r, int c) { return entry(r,c); }; protected: int h; // The height of the matrix. int w; // The width of the matrix. }; #endif gpml-matlab-v3.1-2010-09-27/util/lbfgsb/matlabprogram.cpp000644 000765 000024 00000016452 11422240325 022633 0ustar00carlstaff000000 000000 #include "matlabprogram.h" #include "array.h" // Function definitions. // ----------------------------------------------------------------- // Copy the matrix elements to the destination array. It is assumed // that sufficient memory has been allocated for the destination // array. The copying preserves the column-major ordering of the // matrix elements. void copyElemsFromMatrix (const Matrix& source, double* dest) { Matrix destmatrix(dest,source.height(),source.width()); destmatrix = source; } // Copy the elements from the source array into the matrix. void copyElemsToMatrix (const double* source, Matrix& dest) { dest.inject(source); } // Copy the elements contained in all the matrices to the destination // array. It is assumed that sufficient memory has been allocated for // the destination array. The copying preserves the column-major // ordering of the matrix elements. void copyElemsFromArrayOfMatrices (const ArrayOfMatrices& source, double* dest) { // Repeat for each matrix. for (int i = 0; i < source.length(); i++) { copyElemsFromMatrix(*source[i],dest); dest += source[i]->length(); } } // Copy the elements from the source array into the array of matrices. void copyElemsToArrayOfMatrices (const double* source, ArrayOfMatrices& dest) { // Repeat for each matrix. for (int i = 0; i < dest.length(); i++) { copyElemsToMatrix(source,*dest[i]); source += dest[i]->length(); } } int getBoundType (double& lb, double& ub) { bool haslb = !mxIsInf(lb); bool hasub = !mxIsInf(ub); int btype = haslb + 3*hasub - 2*(haslb && hasub); lb = haslb ? lb : 0; ub = hasub ? ub : 0; return btype; } // Function definitions for class MatlabProgram. // ----------------------------------------------------------------- MatlabProgram::MatlabProgram (ArrayOfMatrices& variables, const ArrayOfMatrices& lowerbounds, const ArrayOfMatrices& upperbounds, const MatlabString* objFunc, const MatlabString* gradFunc, const MatlabString* iterFunc, mxArray* auxData, int m, int maxiter, double factr, double pgtol) : Program(variables.numelems(),0,0,0,0,m,maxiter,factr,pgtol), variables(variables) { x = new double[n]; lb = new double[n]; ub = new double[n]; btype = new int[n]; this->objFunc = objFunc; this->gradFunc = gradFunc; this->iterFunc = iterFunc; // Copy some pieces of information from the input arguments to the // respective program structures. copyElemsFromArrayOfMatrices(variables,x); copyElemsFromArrayOfMatrices(lowerbounds,lb); copyElemsFromArrayOfMatrices(upperbounds,ub); // Set the bound types. for (int i = 0; i < n; i++) btype[i] = getBoundType(lb[i],ub[i]); // Create a Matlab array for the variables. int nv = variables.length(); varInputs = new mxArray*[nv]; varMatlab = new ArrayOfMatrices(varInputs,variables); // Set up the inputs to the objective callback function. numInputsObjFunc = nv + (bool) auxData; inputsObjFunc = new mxArray*[numInputsObjFunc]; copymemory(varInputs,inputsObjFunc,nv); if (auxData) inputsObjFunc[nv] = auxData; // Set up the inputs to the gradient callback function. numInputsGradFunc = numInputsObjFunc; inputsGradFunc = inputsObjFunc; // If necessary, set up the inputs to the iterative callback // function. The first two inputs are the iteration and the function // objective. The next of inputs are the values of the variables // then, optionally, the auxiliary data. if (iterFunc) { numInputsIterFunc = 2 + nv + (bool) auxData; inputsIterFunc = new mxArray*[numInputsIterFunc]; mxArray** ptr = inputsIterFunc; tMatlab = new MatlabScalar(*ptr++,0); fMatlab = new MatlabScalar(*ptr++,0); copymemory(varInputs,ptr,nv); ptr += nv; if (auxData) *ptr = auxData; } } MatlabProgram::~MatlabProgram() { delete[] x; delete[] lb; delete[] ub; delete[] btype; // Deallocate the Matlab arrays. int nv = variables.length(); for (int i = 0; i < nv; i++) mxDestroyArray(varInputs[i]); delete varMatlab; delete[] varInputs; // If necessary, deallocate the additional Matlab arrays that act as // inputs to the iterative callback function. if (iterFunc) { delete fMatlab; delete tMatlab; // delete[] inputsIterFunc; } } double MatlabProgram::computeObjective (int n, double* x) { int nlhs = 1; // The number of outputs from Matlab. mxArray* plhs[nlhs]; // The outputs from the Matlab routine. // Copy the current value of the optimization variables. copyElemsToArrayOfMatrices(x,*varMatlab); // Call the designated Matlab routine for evaluating the objective // function. It takes as input the values of the variables, and // returns a single output, the value of the objective function. if (mexCallMATLAB(nlhs,plhs,numInputsObjFunc,inputsObjFunc,*objFunc)) throw MatlabException("Evaluation of objective function failed in \ call to MATLAB routine"); // Get the result passed back from the Matlab routine. MatlabScalar matlabOutput(plhs[0]); double objective = matlabOutput; // Free the dynamically allocated memory. mxDestroyArray(plhs[0]); return objective; } void MatlabProgram::computeGradient (int n, double* x, double* g) { int nlhs = variables.length(); // The number of outputs from Matlab. mxArray* plhs[nlhs]; // The outputs from the Matlab routine. // Copy the current value of the optimization variables. copyElemsToArrayOfMatrices(x,*varMatlab); // Call the designated Matlab routine for computing the gradient // of the objective function. It takes as input the values of the // variables, and returns as many outputs corresponding to the // partial derivatives of the objective function with respect to // the variables at their curent values. if (mexCallMATLAB(nlhs,plhs,numInputsGradFunc,inputsGradFunc,*gradFunc)) throw MatlabException("Evaluation of objective gradient failed in \ call to MATLAB routine"); // Get the result passed back from the Matlab routine. ArrayOfMatrices matlabOutput((const mxArray**) plhs,nlhs); for (int i = 0; i < variables.length(); i++) if (matlabOutput[i]->length() != variables[i]->length()) throw MatlabException("Invalid gradient passed back from MATLAB \ routine"); copyElemsFromArrayOfMatrices(matlabOutput,g); // Free the dynamically allocated Matlab outputs. for (int i = 0; i < variables.length(); i++) mxDestroyArray(plhs[i]); } void MatlabProgram::iterCallback (int t, double* x, double f) { if (iterFunc) { // Copy the current iteration, the current value of the objective // function, and the current value of the optimization variables. *tMatlab = t; *fMatlab = f; copyElemsToArrayOfMatrices(x,*varMatlab); // Call the Matlab iterative callback routine. Since there are no // outputs, there is no memory to deallocate. if (mexCallMATLAB(0,0,numInputsIterFunc,inputsIterFunc,*iterFunc)) throw MatlabException("Call to MATLAB iterative callback function \ failed"); } } SolverExitStatus MatlabProgram::runSolver() { SolverExitStatus status; // The return value. status = Program::runSolver(); // Copy the solution to the class member "variables". copyElemsToArrayOfMatrices(x,variables); return status; } gpml-matlab-v3.1-2010-09-27/mean/meanConst.m000644 000765 000024 00000001217 11447142622 020111 0ustar00carlstaff000000 000000 function A = meanConst(hyp, x, i) % Constant mean function. The mean function is parameterized as: % % m(x) = c % % The hyperparameter is: % % hyp = [ c ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-08-04. % % See also MEANFUNCTIONS.M. if nargin<2, A = '1'; return; end % report number of hyperparameters if numel(hyp)~=1, error('Exactly one hyperparameter needed.'), end c = hyp; if nargin==2 A = c*ones(size(x,1),1); % evaluate mean else if i==1 A = ones(size(x,1),1); % derivative else A = zeros(size(x,1),1); end endgpml-matlab-v3.1-2010-09-27/mean/meanSum.m000644 000765 000024 00000003511 11447142622 017566 0ustar00carlstaff000000 000000 function A = meanSum(mean, hyp, x, i) % meanSum - compose a mean function as the sum of other mean functions. % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls other mean functions to do the actual work. % % m(x) = \sum_i m_i(x) % % Copyright (c) by Carl Edward Rasmussen & Hannes Nickisch 2010-08-04. % % See also MEANFUNCTIONS.M. for ii = 1:numel(mean) % iterate over mean functions f = mean(ii); if iscell(f{:}), f = f{:}; end % expand cell array if necessary j(ii) = cellstr(feval(f{:})); % collect number hypers end if nargin<3 % report number of parameters A = char(j(1)); for ii=2:length(mean), A = [A, '+', char(j(ii))]; end; return end [n,D] = size(x); v = []; % v vector indicates to which mean parameters belong for ii = 1:length(mean), v = [v repmat(ii, 1, eval(char(j(ii))))]; end if nargin==3 % compute mean vector A = zeros(n,1); % allocate space for ii = 1:length(mean) % iteration over summand functions f = mean(ii); if iscell(f{:}), f = f{:}; end % expand cell array if needed A = A + feval(f{:}, hyp(v==ii), x); % accumulate means end else % compute derivative vector if i<=length(v) ii = v(i); % which mean function j = sum(v(1:i)==ii); % which parameter in that mean f = mean(ii); if iscell(f{:}), f = f{:}; end % dereference cell array if necessary A = feval(f{:}, hyp(v==ii), x, j); % compute derivative else A = zeros(n,1); end endgpml-matlab-v3.1-2010-09-27/mean/meanZero.m000644 000765 000024 00000000603 11422240323 017726 0ustar00carlstaff000000 000000 function A = meanZero(hyp, x, i) % Zero mean function. The mean function does not have any parameters. % % m(x) = 0 % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-01-10. % % See also MEANFUNCTIONS.M. if nargin<2, A = '0'; return; end % report number of hyperparameters A = zeros(size(x,1),1); % derivative and mean gpml-matlab-v3.1-2010-09-27/mean/meanPow.m000644 000765 000024 00000001721 11422240323 017556 0ustar00carlstaff000000 000000 function A = meanPow(mean, hyp, x, i) % meanPow - compose a mean function as the power of another mean function. % % The degree d has to be a strictly positive integer. % % m(x) = m_0(x) ^ d % % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls other mean function to do the actual work. % % Copyright (c) by Carl Edward Rasmussen & Hannes Nickisch 2010-06-18. % % See also MEANFUNCTIONS.M. d = mean{1}; d = abs(floor(d)); d = max(d,1); % positive integer degree mean = mean{2}; if ~iscell(mean), mean = {mean}; end if nargin<3 % report number of parameters A = feval(mean{:}); return end [n,D] = size(x); if nargin==3 % compute mean vector A = feval(mean{:},hyp,x).^d; else % compute derivative vector A = ( d*feval(mean{:},hyp,x).^(d-1) ).* feval(mean{:},hyp,x,i); endgpml-matlab-v3.1-2010-09-27/mean/meanMask.m000644 000765 000024 00000002371 11422240323 017706 0ustar00carlstaff000000 000000 function A = meanMask(mean, hyp, x, i) % Apply a mean function to a subset of the dimensions only. The subset can % either be specified by a 0/1 mask by a boolean mask or by an index set. % % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls another mean function to do the actual work. % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-07-15. % % See also MEANFUNCTIONS.M. mask = mean{1}(:); % either a binary mask or an index set mean = mean(2); % mean function to be masked nh_string = feval(mean{:}); % number of hyperparameters of the full mean if max(mask)<2 && length(mask)>1, mask = boolean(mask); end % convert 1/0->T/F if islogical(mask), D = sum(mask); else D = length(mask); end % masked dimension if nargin<3, A = num2str(eval(nh_string)); return; end % number of parameters if eval(nh_string)~=length(hyp) % check hyperparameters error('number of hyperparameters does not match size of masked data') end if nargin==3 A = feval(mean{:}, hyp, x(:,mask)); else % compute derivative A = feval(mean{:}, hyp, x(:,mask), i); endgpml-matlab-v3.1-2010-09-27/mean/meanLinear.m000644 000765 000024 00000001330 11447142622 020231 0ustar00carlstaff000000 000000 function A = meanLinear(hyp, x, i) % Linear mean function. The mean function is parameterized as: % % m(x) = sum_i a_i * x_i; % % The hyperparameter is: % % hyp = [ a_1 % a_2 % .. % a_D ] % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-01-10. % % See also MEANFUNCTIONS.M. if nargin<2, A = 'D'; return; end % report number of hyperparameters [n,D] = size(x); if any(size(hyp)~=[D,1]), error('Exactly D hyperparameters needed.'), end a = hyp; if nargin==2 A = x*a; % evaluate mean else if i<=D A = x(:,i); % derivative else A = zeros(n,1); end endgpml-matlab-v3.1-2010-09-27/mean/meanOne.m000644 000765 000024 00000000746 11447142622 017552 0ustar00carlstaff000000 000000 function A = meanOne(hyp, x, i) % One mean function. The mean function does not have any parameters. % % m(x) = 1 % % Copyright (c) by Carl Edward Rasmussen and Hannes Nickisch, 2010-08-04. % % See also MEANFUNCTIONS.M. if nargin<2, A = '0'; return; end % report number of hyperparameters if nargin==2 A = ones(size(x,1),1); % evaluate mean else A = zeros(size(x,1),1); % derivative endgpml-matlab-v3.1-2010-09-27/mean/meanProd.m000644 000765 000024 00000003754 11447142622 017737 0ustar00carlstaff000000 000000 function A = meanProd(mean, hyp, x, i) % meanProd - compose a mean function as the product of other mean functions. % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls other mean functions to do the actual work. % % m(x) = \prod_i m_i(x) % % Copyright (c) by Carl Edward Rasmussen & Hannes Nickisch 2010-08-04. % % See also MEANFUNCTIONS.M. for ii = 1:numel(mean) % iterate over mean functions f = mean(ii); if iscell(f{:}), f = f{:}; end % expand cell array if necessary j(ii) = cellstr(feval(f{:})); % collect number hypers end if nargin<3 % report number of parameters A = char(j(1)); for ii=2:length(mean), A = [A, '+', char(j(ii))]; end; return end [n,D] = size(x); v = []; % v vector indicates to which mean parameters belong for ii = 1:length(mean), v = [v repmat(ii, 1, eval(char(j(ii))))]; end A = ones(n,1); % allocate space if nargin==3 % compute mean vector for ii = 1:length(mean) % iteration over summand functions f = mean(ii); if iscell(f{:}), f = f{:}; end % expand cell array if needed A = A.*feval(f{:}, hyp(v==ii), x); % accumulate means end else % compute derivative vector if i<=length(v) ii = v(i); % which mean function j = sum(v(1:i)==ii); % which parameter in that mean for jj = 1:length(mean) f = mean(jj); if iscell(f{:}), f = f{:}; end % dereference cell array if necessary if jj==ii A = A .* feval(f{:}, hyp(v==jj), x, j); % multiply derivative else A = A .* feval(f{:}, hyp(v==jj), x); % multiply mean end end else A = zeros(n,1); end endgpml-matlab-v3.1-2010-09-27/mean/meanScale.m000644 000765 000024 00000001575 11422240323 020047 0ustar00carlstaff000000 000000 function A = meanScale(mean, hyp, x, i) % meanScale - compose a mean function as a scaled version of another one. % % m(x) = a * m_0(x) % % The hyperparameter is: % % hyp = [ a ] % % This function doesn't actually compute very much on its own, it merely does % some bookkeeping, and calls other mean function to do the actual work. % % Copyright (c) by Carl Edward Rasmussen & Hannes Nickisch 2010-07-15. % % See also MEANFUNCTIONS.M. if nargin<3 % report number of parameters A = [feval(mean{:}),'+1']; return end [n,D] = size(x); a = hyp(1); if nargin==3 % compute mean vector A = a*feval(mean{:},hyp(2:end),x); else % compute derivative vector if i==1 A = feval(mean{:},hyp(2:end),x); else A = a*feval(mean{:},hyp(2:end),x,i-1); end end