Survey Propagation beyond Constraint Satisfaction Problems

Survey propagation (SP) is a message passing procedure that attempts to model all the fixed points of Belief Propagation (BP), thereby improving BP’s approximation in loopy graphs where BP’s assumptions do not hold. For this, SP messages represent distributions over BP messages. Unfortunately this requirement makes SP intractable beyond constraint satisfaction problems because, to perform general SP updates, one has to operate on distributions over a continuous domain. We propose an approximation scheme to efficiently extend the application of SP to marginalization in binary pairwise graphical models. Our approximate SP has O(DK log(DK)τ) complexity per iteration, where τ is the complexity of BP per iteration, D is the maximum node degree and K is a resolution constant controlling the approximation’s fidelity. Our experiments show that this method can track many BP fixed points, achieving a high marginalization accuracy within a few iterations, in difficult settings where BP is often non-convergent and inaccurate.