Learning of rules in an expert system with a probabilistic expert

The authors deal with a theory of learning in expert systems. An environment is considered in which the expert is susceptible to the process of learning. To describe the state of such an expert during the formulation of rules, a notion of a human expert in learning phase is introduced. The modeling of knowledge is done by condition-action-type probabilistic rules. The forms as well as the parameters of these rules are furnished by the expert. An algorithm is presented for automatically learning the parameters of the rules with the help of a probabilistic expert. This algorithm is computationally feasible, and the results of simulation show the expedient learning of parameters of the rules.<>