 |
 |
Dear all, my question is simple: can the Free Energy, the objective function maximized by Variational Bayes, be positive? Since it is a lower bound to the log-model evidence, my guess is that it cannot be positive as the log of a probability is negative. Nevertheless, I am sometimes obtaining positive values and I do not know if there is a problem in my code, if there is a numerical issue due to round off or if actually the Free Energy could be positive. To note, the scheme I am using is the one proposed here, http://dx.doi.org/10.1109/TSP.2008.2005752, with conjugate prior and thus without Laplace approximation. Any suggestion will be much appreciated! Matteo Tonietto Department of Information Engineering University of Padova, Italy http://fair.dei.unipd.it/matteo-tonietto/