Reasoning with incomplete information in a multivalued multiway causal tree using the maximum entropy formalism

Abstract

Expert systems that use causal probabilistic networks require the user to supply complete causal information regarding the causal probabilities to be used. This paper describes a method using the maximum entropy formalism that enables such expert systems to operate with incomplete causal information for certain classes of causal networks. It has been shown that, in the general case, solving causal networks using maximum entropy techniques is NP-complete. However, we show that for multivalued causal multiway trees—a nontrivial class of causal networks—the problem of estimating missing information is only linear. © 1998 John Wiley & Sons, Inc.