Relevance in probabilistic models: "Backyards" in a "small World"

Authors:

Marek J. Druzdzel
University of Pittsburgh
Department of Information Science
e-mail: marek@sis.pitt.edu

Henri J. Suermondt
Hewlett-Packard Laboratories
e-mail: suermondt@hpl.hp.com

Abstract:

Each of the variables in a large probabilistic model may be relevant for some types of reasoning within this model, but rarely will all of them participate in reasoning related to a single query. We review a variety of schemes to identify variables that given certain observations are relevant to a query of interest.