Evidential support logic, FRIL and case based reasoning

✍ Scribed by J. F. Baldwin

• Publisher: John Wiley and Sons
• Year: 1993
• Tongue: English
• Weight: 942 KB
• Volume: 8
• Category: Article
• ISSN: 0884-8173
• DOI: 10.1002/int.4550080904

✦ Synopsis

Knowledge representation for expert systems and decision support systems is often in the form of rules and answering queries is performed by backward and forward reasoning. Case based reasoning is an alternative to this. In its most basic form, a query is answered by reference with data given as part of the query to similar cases in a data base. In this article we will use a mixture of case based reasoning and rule inference methods. Rules will represent generalizations of information relevant to prototypical cases. The prototypical cases are chosen from a database of examples. Partial matchings of features in a rule are used to infer a conclusion even when information to evaluate the body of the rule is incomplete. Features are given importance weights and inferred conclusions from various rules are combined using mass assignment theory 0 1993 John Wiley & Sons. Inc.

I. INTRODUCTION

Knowledge representation for expert systems and decision support systems is often in the form of rules and answering queries is performed by backward and forward reasoning. For application to control see Refs. I and 2.

Rules take the form of head IF body : p where p is optional. If p is given this is interpreted as Pr(head1body) = p . If no p is given then the rule is interpreted as body 3 head. The body is of the conjunctive form headl, . . . , headn where headi is either a head of another rule or a fact in the knowledge base.

The truth or probability of the body is determined using the logic or probability rules for conjunction which ever is appropriate.

We can generalize this further by allowing for p to be an interval containing the probability of the head given the body. This interval, [ a , p ] , is called a *Professor J. F. Baldwin is a SERC Senior Research Fellow.