Implementing Belief Propagation on P-Grid
Nicolas Bonvin, Grégoire Montavon, Damien Auroux · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · 2007
Abstract—Some applications require to solve a Bayesian network in a distributed fashion. It can lead to a large amount of messages exchanged between hosts if the variables of the Bayesian network are not allocated to good hosts. The technique we implement uses an algorithm that allocates variables to hosts so that the tension of the Bayesian network is minimized. We present and discuss our results. I.