Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!mit-eddie!bbn.com!BKort From: BKort@bbn.com (Barry Kort) Newsgroups: comp.ai Subject: Re: Fuzzy Logic, Anyone? Message-ID: <59997@bbn.BBN.COM> Date: 12 Oct 90 15:15:19 GMT References: <69385@lll-winken.LLNL.GOV> Sender: news@bbn.com Organization: BBN Labs Lines: 41 In article <69385@lll-winken.LLNL.GOV> loren@tristan.llnl.gov (Loren Petrich) writes: > Has anyone around here used Fuzzy Logic? > > My understanding of it is that, in it, predicates may not only > be "true" or "false" but anywhere in between. If 0 is false and 1 is > true, then the truth-value of a predicate will lie between 0 and 1 > inclusive of the two limits. > > It turns out that one can define the operations "not", "and" > and "or"; one can even define several pairs of the latter two > operations. A few years ago, Debby Guerrera and I used continuous-valued logic in precisely the way you outline above as part of a prototype diagnostic expert system. The diagnostic subsystem entertained a collection of hypotheses regarding possible fault conditions in a complex system. Evidence for and against each hypothesis was accumulated and integrated into a "degree of belief". If a particular fault condition was indeed present, the preponderance of confirmatory evidence would eventually push the degree of belief close enough to 1 to justify a postiive diagnosis. We built the underlying model in Smalltalk and used Prolog for the expert system. In retrospect, there was no real need to use Prolog to do the diagnostic reasoning, but we wanted to learn more about rule-based and goal-directed expert systems. There is nothing inherently msyterious or difficult about continuous-valued logic. The real challenge is to recognize the tell-tale clues from the observable behavior of the system when various fault conditions are present, and weave those clues into the calculations. After that, one has to run the model for extended periods to prove that that the diagnostic system can reliably detect all reasonable combinations of fault conditions. Barry Kort Visiting Scientist BBN Labs Cambridge, MA