Fuzzy Calculus by RBF Neural Networks
The paper presents novel modeling of fuzzy inference system by using the ‘fuzzified’ radial basis function (RBF) neural network (NN). RBF NN performs the mapping of the antecedent fuzzy numbers (a.k.a. membership functions, attributes, possibilities degrees) into the consequent ones. In this way, an RBF NN is capable of performing the rigorous calculus with fuzzy numbers. Prior the mapping, both the antecedents and the consequents are discretized and transferred into the n-dimensional and m-dimensional ‘fuzzy’ vectors. These vectors present the training inputs and outputs of an RBF NN and, in this way, an RBF network performs an exact R n → R m mapping. The generalization capacity of such a neural implementation is superior to the ability of the original fuzzy model.
KeywordsMembership Function Radial Basis Function Fuzzy Number Fuzzy Model Inference Rule
Unable to display preview. Download preview PDF.
- Ishibuchi, H. Okada and H. Tanaka. Fuzzy neural networks with fuzzy weights and fuzzy biases. Proc. ICNN’93, San Francisco pp., 1650–1655, 1993b.Google Scholar
- Kecman V., Pfeiffer B-M., ‘Exploiting the Structural Equivalence of Learning Fuzzy Systems and Radial Basis Function Neural Networks’, 2-nd Europ. Congr. on Intell. Techn. and Soft Comput., EUFIT ‘84, Aachen, Vol. 1, pp. 58–66, 1994.Google Scholar
- Li, Z.Q., Kecman, V., Ichikawa, A., ‘Fuzzified Neural Network Based on Fuzzy Number Operations’, Fuzzy Sets and Systems, (to appear), 2002.Google Scholar