Fast Learning Algorithm for Fuzzy Inference Systems using Vector Quantization

It is known that learning methods of fuzzy inference systems using vector quantization (VQ) and steepest descend method (SDM) are superior in terms of the number of rules. However, they need a great deal of learning time. The cause could be that both of VQ and SDM perform only local searches. On the other hand, it has been shown that a learning method of radial basis function (RBF) networks using VQ and generalized inverse method (GIM) is much fast. In this paper, we propose a new learning method using VQ, GIM and SDM. The method iterates three stages in the outer loop of the algorithm. The first stage adjust the fuzzy rule arrangement by using VQ, the second one determines the weights of fuzzy rules by using GIM, and the third one updates both of the rule arrangement and the weights. In order to demonstrate the validity of the proposed method, numerical simulations for function approximation and pattern classification problems are performed. Specifically, it is shown that the proposed method reduces the learning time to about one-tenth compared to conventional methods in function approximation problem.