An Improved LAZY-AR Approach to Bayesian Network Inference

We propose LAZY arc-reversal with variable elimination (LAZY-ARVE) as a new approach to probabilistic inference in Bayesian networks (BNs). LAZY-ARVE is an improvement upon LAZY arcreversal (LAZY-AR), which was very recently proposed and empirically shown to be the state-of-the-art method for exact inference in discrete BNs. The primary advantage of LAZY-ARVE over LAZY-AR is that the former only computes the actual distributions passed during inference, whereas the latter may perform unnecessary computation by constructing irrelevant intermediate distributions. A comparison between LAZYAR and LAZY-ARVE, involving processing evidence in a real-world BN for coronary heart disease, is favourable towards LAZY-ARVE.

• Type: Conference
• Year: 2006
• Where: AI
• Authors: Cory J. Butz, Shan Hua
• Keywords: AI 2006, Artificial Intelligence, Exact Inference, Irrelevant Intermediate Distributions, Probabilistic Inference