A0737
Title: Exchangeable random permutations for Bayesian graph matching
Authors: Nathaniel Josephs - North Carolina State University (United States) [presenting]
Abstract: The graph-matching problem is a classic task that involves finding the correspondence between the vertices of two graphs. A new class of nonparametric priors is introduced for permutations by borrowing ideas from the extensive literature on partition structures. This enables a Bayesian approach to graph matching that combines the position-aware Chinese restaurant process with a correlated stochastic block model likelihood. A node-wise blocked Gibbs sampler is proposed for posterior inference, as well as an efficient posterior summary technique that leverages variation-of information (VI) summaries for partitions.
