Methods of revision principle for approximate reasoning

Compositional inference and compatibility-modification inference are two main approaches for approximate reasoning (Baldwin, 1979a; 1979b; Dubois, 1985; Mizumoto, 1981; Mizumoto, 1987; Tsukamoto, 1979; Yager, 1980; Zadeh, 1975a; 1975b; 1979). The former realizes inference by obtaining an implication relation between antecedent and consequent of a rule and then composing the input with the relation (Zadeh, 1975a). The latter realizes inference by determining the measure of satisfaction between input and antecedent of a rule and then using the measure to modify the rule's consequent (Dubois, 1985). The revision principle was proposed in a different way: it is under such a belief that the modification (revision) of consequent should be caused only by the difference (deviation) between input (given fact) and antecedent. In other words, when a method of revision principle is used to approximate reasoning the consequent will always be obtained as output if input is the same as the antecedent. The revising processing is based on some kind of relation between antecedent and consequent, which can be linear relation or semantic relation. We introduce five revising methods and then evaluate them by relation keeping property.