Relevance of Evidence in Bayesian Networks

Abstract

AbstractFor many inference tasks in Bayesian networks, computational efforts can be restricted to a relevant part of the network. Researchers have studied the relevance of a network’s variables and parameter probabilities for such tasks as sensitivity analysis and probabilistic inference in general, and identified relevant sets of variables by graphical considerations. In this paper we study relevance of the evidence variables of a network for such tasks as evidence sensitivity analysis and diagnostic test selection, and identify sets of variables on which computational efforts can focus. We relate the newly identified sets of relevant variables to previously established relevance sets and address their computation compared to these sets. We thereby paint an overall picture of the relevance of various variable sets for answering questions concerning inference and analysis in Bayesian network applications.KeywordsBayesian Network ApplicationsGraphical ConsiderationsSensitivity SetEvidence NodesIrrelevant NodesThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.