This paper is concerned with fuzzy modeling in some reductionmethods of inference rules with gradient descent. Reduction methodsare presented, which have a reduction mech- anism of the rule unitthat is applicable in three parameters-the central value and thewidth of the membership function in the an- tecedent part, and thereal number in the consequent part-which constitute the standardfuzzy system. In the present techniques, the necessary number ofrules is set beforehand and the rules are sequentially deleted to theprespecified number.
展开▼