A framework for comparative analysis of belief revision models in rule-based systems

Abstract

Belief revision in standard rule-based systems is NP-hard, and it is necessary to resort to quasiprobabilistic belief languages which are quite problematic on normative grounds. The author presents a framework designed to study the Bayesian interpretation and cognitive appeal of these languages. He uses this framework to review some well-known and some recent findings, and comments on the merit of alternative approaches to belief revision in rule-based systems.<>