A fuzzy neural network architecture for fuzzy control and classification
B. Chen, L. L. Hoberock · 2002
In this paper, a fuzzy neural network architecture with fuzzy weights, termed FUZAMP, is proposed to process non-singleton fuzzy data. FUZAMP can also be applied to singleton fuzzy data, namely analog and binary data. For such singleton data, FUZAMP is consistent with fuzzy ARTMAP introduced by Carpenter and Grossberg et al. (1991), although fuzzy ARTMAP cannot handle fuzzy data. FUZAMP consists of a fuzzified version of ART and a fuzzified mapping network. The fuzzified mapping network enables FUZAMP to learn fuzzy rules from fuzzy input data. As an extension of fuzzy ARTMAP, concepts of fuzzy norm, fuzzy subset function, fuzzy match tracking, fuzzy resonance, and fuzzy complement coding are presented and defined for FUZAMP. FUZAMP can be trained for on-line supervised learning of fuzzy and crisp recognition categories from singleton fuzzy data, non-singleton fuzzy data, or combined singleton and non-singleton fuzzy data, and it has superior performance with fast convergence and guaranteed stability for fuzzy data.