Join tree propagation with prioritized messages

Abstract

Current join tree propagation algorithms treat all propagated messages as being of equal importance. On the contrary, it is often the case in real-world Bayesian networks that only some of the messages propagated from one join tree node to another are relevant to subsequent message construction at the receiving node. In this article, we propose the first join tree propagation algorithm that identifies and constructs the relevant messages first. Our approach assigns lower priority to the irrelevant messages as they only need to be constructed so that posterior probabilities can be computed when propagation terminates. Experimental results, involving the processing of evidence in four real-world Bayesian networks, empirically demonstrate an improvement over the stateof-the-art method for exact inference in discrete Bayesian networks.

d-Separation: Strong Completeness of Semantics in Bayesian Network Inference

“…The unity-potential 1(v i ) for v i is a function 1 mapping every element of dom(v i ) to one. The unitypotential for a non-empty set X = {v 1 …”

“…Here, X and E are disjoint subsets of U , and E is observed taking value e. We describe a basic algorithm for computing p(X|E = e), called variable elimination (VE), first put forth in [17]. We do not consider alternative approaches to inference such as conditioning [6] and join tree propagation [1,2,10]. Inference involves the elimination of variables.…”

Fast Factorisation of Probabilistic Potentials and Its Application to Approximate Inference in Bayesian Networks

“…Several exact 2,3,4,5,6,7,8 and approximate 9,10,11,12,13 inference algorithms can be found in the literature, due to the fact that it is an NP-hard problem, 14,15 which justifies the study of new techniques and algorithms with the aim of widening the class of affordable problems. Some of the most relevant inference algorithms incorporate the ability of dealing with factorised representations of the potentials that represent the probabilistic information.…”

Answering queries in hybrid Bayesian networks using importance sampling

“…A review of recent literature shows the variety of applications in which they have been successfully used [1,10,24,35,45]. One of the main reasons for using them as the inference engine in a decision support system is that efficient reasoning algorithms can be designed, taking advantage of their structure [2,3,16,44,43,30,29].…”