Fast Convergent Algorithms for Expectation Propagation Approximate Bayesian Inference

TL;DR

This paper introduces a new, provably convergent algorithm for expectation propagation in Bayesian inference for continuous-variable models, significantly improving computational speed over existing methods.

Contribution

It presents a novel, convergent EP algorithm that combines ideas from previous works and achieves at least ten times faster performance.

Findings

Algorithm is provably convergent.

Runs at least an order of magnitude faster than standard EP.

Applicable to continuous-variable graphical models.

Abstract

We propose a novel algorithm to solve the expectation propagation relaxation of Bayesian inference for continuous-variable graphical models. In contrast to most previous algorithms, our method is provably convergent. By marrying convergent EP ideas from (Opper&Winther 05) with covariance decoupling techniques (Wipf&Nagarajan 08, Nickisch&Seeger 09), it runs at least an order of magnitude faster than the most commonly used EP solver.