A fuzzy-ART-enhanced neural classifier
Jui‐Sheng Chou, C.-S. Ho · 2003
A neural network classifier enhanced by the fast-learning fuzzy ART mechanism is presented. The learning algorithm combines the fuzzy ART procedure for structure learning and the error backpropagation learning scheme for parameter learning. The former helps self-generate a proper hidden layer and auto-initialize the connection weights of related layers to alleviate the local minima problem and to improve the learning speed. The experiments show that the proposed model can exhibit high-quality classification capability in either a continuous, discrete, linear or a nonlinear domain.