Meta-Level Control Through Fault Detection and Diagnosis

Book One

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 4

Automated Reasoning

Control strategies in most complex problem-solving systems, though highly parameterized, are not adaptive to the characteristics of the particular task being solved. If the characteristics of the task are atypical, a fixed control strategy may cause incorrect or inefficient processing. We present an approach for adapting the control strategy by introducing a meta-level control component into the problem-solving architecture. This meta-level control component is based on the paradigm of Fault Detection/Diagnosis. Our presentation will concentrate on modeling the problem-solving system and on the inference techniques necessary to use this model for diagnosis. We feel that meta-level control based on the Fault Detection/Diagnosis paradigm represents a new approach to introducing more sophisticated control into a problem- solving system.