function d = mvn_div_js(m1, m2, use_kl)
%MVN_DIV_JS Compute the Jensen-Shannon (JS) divergence between two multivariate normals.
% [d] = MVN_DIV_JS(m1, m2) computes the JS divergence between two
% multivariate normals. d is never negative.
% The JS divergence is defined as:
% d = 0.5*KL(m1, m1+m2) + 0.5*KL(m2, m1+m2)
% (c) 2010-2011, Dominik Schnitzer, <firstname.lastname@example.org>
% This file is part of the MVN Octave/Matlab Toolbox
% MVN is free software: you can redistribute it and/or modify
% it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or
% (at your option) any later version.
% MVN is distributed in the hope that it will be useful,
% but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
% GNU General Public License for more details.
% You should have received a copy of the GNU General Public License
% along with MVN. If not, see <http://www.gnu.org/licenses/>.
% m12 = mvn_bregmancentroid_kl_left([m1 m2]);
% using Cholesky & more Optimization
m12.m = 0.5*m1.m + 0.5*m2.m;
m12.cov = 0.5*(m1.cov + m1.m*m1.m') + 0.5*(m2.cov + m2.m*m2.m') ...
m12_chol = chol(m12.cov);
m12.logdet = 2*sum(log(diag(m12_chol)));
if ((nargin > 2) && (use_kl == 1))
m12_ui = m12_chol\eye(length(m12.m));
m12.icov = m12_ui*m12_ui';
d = 0.5*mvn_div_kl(m1, m12) + 0.5*mvn_div_kl(m2, m12);
% Speedup original (entropy):
% d = mvn_entropy(m12) - 0.5*mvn_entropy(m1) - 0.5*mvn_entropy(m2);
d = 0.5*m12.logdet - 0.25*m1.logdet - 0.25*m2.logdet;
d = max(d, 0);