Representing Diagnostic Knowledge for Probabilistic Horn Abduction

This paper presents a simple logical framework for abduction, with probabilities associated with hypotheses. The language is an extension to pure Prolog, and it has straight-forward implementations using branch and bound search with either logic-programming technology or ATMS technology. The main focus of this paper is arguing for a form of representational adequacy of this very simple system for diagnostic reasoning. It is shown how it can represent model-based knowledge, with and without faults, and with and without non-intermittency assumptions. It is also shown how this representation can represent any probabilistic knowledge representable in a Bayesian belief network.