Maintenance in Probabilistic Knowledge-Based Systems

Recent developments using direc.ted a.C"yclical graphs (i .e., influence diagrams and Bayesian networks) for knowledge representation have lessened the problems of using probability in knowledge-based systems (KBS). 11ost current research involves the efficient propagation of new evidenee, but, lit.tle has been done concerning the maint.e­ nance of domain-specific knowledge, which includes the probabilistic information about the problem domain. By making use of conditional independencies represented in the graphs, however, probability assessments are required only for certain variables when the knowledge base is updated. The purpose of this study was to invest.igate, for those variables which require probability assessments, ways to reduce the amount of new knowledge required from. the expert when updating probabilistic information in a probabilistic knowledgebased system. Three special .cases (ignored outcome, split. out<'ome, and assumed con:;t.a.nt outcome) were identified under which many of the onginal probabilities (t.hose already in the knowledge-base) do not need to be reas:;essed when maintenance is required.