Fuzzy logic

At bottom, a data bit in a computer is “all or nothing” (1 or 0). Most decisions in computer code are also all or nothing: Either a condition is satisfied, and execution takes one specified path, or the condition is not satisfied and it goes elsewhere. In real life, of course, many situations fall between the cracks. For example, a business might want to treat a credit applicant who almost qualifies for “A” status different from one who barely made “B.” While a program could be refined to include many gradations between B and A, another approach is to express the degree of “closeness” (or certainty) using fuzzy logic.

In 1965, mathematician L. A. Zadeh introduced the con-cept of the fuzzy set. In a fuzzy set, a given item is not simply either a member or not a member of a specified set. Rather, there is a degree of membership or “suitability” somewhere between 0 (definitely not a member) and 1 (defi-nitely a member). A program using fuzzy logic must include a variety of rules for determining how much certainty to assign in a given case. One way to create rules is to ask experts in a given field (such as credit analysis) to articulate the degree of certainty or confidence they would feel in a given set of circumstances. For physical systems, data can also be correlated (such as the relationship of temperature to the likelihood of failure of a component) and used to cre-ate a rule to be followed by, for example, a chemical process control system.

Fuzzy logic is particularly applicable to the creation of programs (see expert system) that are better able to cope with uncertainty and the need to weigh competing factors in coming to a decision. It can also be used in engineer-ing to allow designers to specify which factors they want to tightly constrain (such as for safety reasons) and which can be allowed more leeway. The system can then come up with optimized design specifications. Fuzzy logic has also been applied to areas such as pattern recognition and image analysis where a number of uncertain observations must often be accumulated and a conclusion drawn about the overall object.