Commonsense Reasoning with Rules, Cases, and Connectionist

Models: A Paradigmatic Comparison

Ron Sun

Department of Computer Science

The University of Alabama

Tuscaloosa, AL 35487

March 30, 1995

Abstract
The paper attempts to explore high-level connectionist models for approximate commonsense reasoning from a broad perspective, investigating their connections to two other prominent paradigms: rule-based reasoning and case-based reasoning. High-level connectionist models, especially the CONSY- DERR architecture, are studied in light of these paradigms. Four criteria are hypothesized regarding these models, and it is concluded that models for approximate commonsense reasoning not only must take into account the conceptual and subconceptual distinction but also make use of the interaction between the two. A related issue is that such models should take into account realistic substrate constraints.

KEYWORDS: hybrid systems, approximate reasoning, features, similarity, fuzzy evidential logic, substrates