Adaptation problems in expert systems

Abstract

Expert system (ES) theory problems have been traditionally considered without any essential connection with common control problems. However, nowadays, because of progress in ES theory development in practice and because of its wide use for solving complex control problems, new items in this field of research are being stressed, many of which are connected with traditional automation control. The combined study of adaptation and ES problems makes it possible to draw significant practical conclusions for systems of both types. In this work an analysis of general problems and of possible approaches to solving them in some special types of ESs based on Bayes inference is carried out. The main attention is focused on such ES characteristics as: (i) inference stability (with respect to variations in probabilistic approximation measures of rule truth); (ii) knowledge base adaptation (rule modification and new rule creation or approximation measure rule tranformation); (iii) knowledge base consistency maintenance.