Cutset Sampling for Bayesian Networks

TL;DR

This paper introduces cutset sampling, a novel method for Bayesian networks that combines sampling and exact inference to improve convergence and efficiency, especially when the network structure allows for bounded induced width.

Contribution

It proposes a new sampling technique that exploits network structure and Rao-Blackwellisation, providing an efficient and scalable alternative to traditional sampling methods.

Findings

Empirical results show improved convergence over standard methods.

Efficient implementation when the network's induced width is bounded.

Demonstrated benefits on various benchmark networks.

Abstract

The paper presents a new sampling methodology for Bayesian networks that samples only a subset of variables and applies exact inference to the rest. Cutset sampling is a network structure-exploiting application of the Rao-Blackwellisation principle to sampling in Bayesian networks. It improves convergence by exploiting memory-based inference algorithms. It can also be viewed as an anytime approximation of the exact cutset-conditioning algorithm developed by Pearl. Cutset sampling can be implemented efficiently when the sampled variables constitute a loop-cutset of the Bayesian network and, more generally, when the induced width of the networks graph conditioned on the observed sampled variables is bounded by a constant w. We demonstrate empirically the benefit of this scheme on a range of benchmarks.