Resuscitation of certainty factors in expert networks
Susan I. Hruska, D.C. Kuncicky, Robert C. Lacher · 1991
An expert network is a form of neural network which captures the rule-based knowledge of an expert system in digraph form. Connectionist learning techniques have been developed for these hybrid systems. In the present work, a study of the performance of these training algorithms in recovering certainty factors for expert networks is presented. Results are reported for the Wine Advisor testbed, a well-known expert system rule base in M.1. Issues in active sampling and learning parameter selection are also discussed.>