Hitachi, Ltd.

Ryuko Someya

Michitaka Kosaka

Hiroshi Yamada

Motohisa Funabashi

Akira Maeda

Toshiro Sasaki

Chizuko Yasunobu

Toshihiko Nakano

Hirotaka Mizuno

Seiji Yasunobu

Fumihiko Mori

Satoru Suemitsu

Yoshikane Sugimura

Fuzzy knowledge stored in a knowledge base is written into a memory. When fuzzy reasoning is executed by a fuzzy reasoning unit based on the content of the memory, a control unit in a central processing unit may modify the knowledge base in the memory. The control unit is programmed such that after it has modified the knowledge base in the memory, it causes a reasoning execution unit to repeatedly and automatically execute predetermined fuzzy reasoning by the number of times corresponding to the number of input data. The control unit also modifies the knowledge base and causes the reasoning execution unit to execute the fuzzy reasoning. By reading out the knowledge base in the memory, it is possible to verify the validity of the fuzzy knowledge without executing the simulation of the fuzzy reasoning. The fuzzy reasoning course of the reasoning execution unit may be stored in another memory. In this case, the control unit may display the fuzzy reasoning result on a display device in an expression easily understandable by a user, based on the content of the other memory.

Method and apparatus for fuzzy knowledge reasoning

In a learning method, training data and verification data each including a set of input and output values are taken as teacher data, and a change of an error between an output value calculated by use of the input data of the verification data and the output value of the verification data and a change of an error between an output value calculated by use of the input value of the training data and the output value of the training data are simultaneously displayed, thereby preventing an overlearning and reducing the number of times of learning. Also in the learning method, fuzzy knowledge data is inputted and displayed, a range of the object of learning is interactively designated from among the displayed fuzzy knowledge, and a learning is made on the basis of a predetermined learning algorithm to change the fuzzy knowledge in the designated range, thereby performing the learning of a fuzzy knowledge with a user's intention reflected.

Learning method and apparatus

A fuzzy evaluation and modification system evaluates and modifies the fuzzy knowledge in a fuzzy reasoning system. The fuzzy knowledge includes fuzzy rules, each described in an 'if . . . , then . . .' format, and membership functions defining meanings of propositions described in the 'if . . . , then' parts. Results of the fuzzy reasoning are monitored and compared with real data of a plurality of real test cases to evaluate the fuzzy knowledge. An influence power of each of the fuzzy rules with respect to the fuzzy reasoning results is attained for all the cases. Superpositional display of real data on the membership functions of the 'if . . .' part and reasoning results on the membership functions of the 'then . . .' part indicates a relationship between the membership function and the real data. The system includes units for obtaining a grade of rule representing adequacy of the fuzzy rule based on the membership functions associated with the 'if' part, attaining defuzzification values of the membership functions related to the 'then' part and obtaining differences between the defuzzification values and the real data. A relationship between the grade of rule and the difference thus obtained between the defuzzification values and the real data is outputted to an output unit. The fuzzy knowledge is modified by an operator in association with the output.

Method and system for evaluating and modifying fuzzy knowledge

Growing community of inventors

Kawasaki, Japan

Ryuko Someya