Reasoning and learning method for fuzzy rules using neural networks and its application to medical diagnosis
Takumi Ichimura, E. Tazaki, Kunihiro Yoshida · 1994
This paper presents the reasoning and the learning method of fuzzy rules using structure level adaptation of neural networks. In a normal neural network's mechanism, during processing of rules, one can observe the following behaviors: case 1) If a neural network does not have enough neurons to be satisfied to infer, then the input weight vector will tend to fluctuate greatly even after a certain period of the learning process. Case 2) If a neural network has enough neurons to infer and even if the input weight vector of each neuron will converge to a certain value, one will be able to turn out unnecessary neurons from the network in the calculation. By observing such behaviors, one can generate or annihilate the specified neurons respectively to achieve an overall good system. In proposed method, the authors give a procedure to derive the neuron generation/annihilation automatically and apply the procedure to a learning system in which the experimental data related to hepatobiliary disorders were used. After learning by using 107 data chosen randomly from all database, the proposed system correctly diagnosed 93% of test data.>