The Role of Tuning Uncertain Inference Systems

This study examined the effects of "tuning" the parameters of the incremental function of MYCIN, the independent function of PROSPECTOR, a probability model that assumes independence, and a simple additive linear equation. The parameters of each of these models were optimized to provide solutions which most nearly approximated those from a full probability model for a large set of simple networks. Surprisingly, MYCIN, PROSPECTOR, and the linear equation performed equivalently; the independence model was clearly more accurate on the networks studied. 1.0 INTRODUCTION A handful of researchers in recent years have attempted to compare various uncertain inference formalisms found in the Artificial Intelligence (AI) literature to probability theory. Some of these papers are apparently intended to provide additional justification for the ad hoc parameters used by some of these formalisms. Heckerman (1], for example, shows that the equations that define MYCIN's certainty factors can be translated into probabilistic terms. Other studies, however, have attempted to use the answers provided by probability theory as a norm against which the accuracy of heuristic formalisms can be measured [2,3,4]. These studies differed somewhat in implementation, but each began with example inference networks. Next, new values were assigned to the evidence nodes, as though additional information were being supplied by a user. Finally, conclusion node values were calculated which reflected the new information, according to the heuristic formalisms under consideration and also according to a probablistic method which provided the minimum *This research was conducted under the McDonnell Douglas Independent Research and Development program.