Reasoning for Inconsistency-- A Taxonomy and a Connectionist Approach
Gadi Pinkas · Open Scholarship Institutional Repository (Washington University in St. Louis) · 1991
A consequence relation (CR) relates sets of beliefs to the appropriate conclusions that might be deduced. Of special interest to Artificial Intelligence are CRs that cope with inconsistency within the set of beliefs. Default reasoning, belief revision, social choice and reasoning from conflicting knowledge sources are just a few examples of mechanisms that need to handle inconsistency, In this paper we show a taxonomy in which many existing mechanisms are mapped, and new interesting ones are revealed. We identify simple relations among the CRs and give a language for their specification. We then show that a large portion of the CRs described by the language can be implementable in neural networks like Boltzman machines and Hopfield nets. The result demonstrates the flexibility of these connectionist models for the approximation of a variety of knowledge level theories.