Z

Previously, we have considered belief propagation (BP) as a way to compute inference for a graph model. Namely, given p(x) and a particular graph model, we want to find the marginal p(xi) for some node i. We will consider BP from the view of statistical physics in this lecture. Recall some concept in statistical physics, for a system with a number of states with energy E(x), the probability of state x under Boltzmann statistics will be p(x) = e− E(x) kT