Study on Adjustment of Learning Rate and Its Application of ART2

ART2 proposed by Carpenter and Grossberg is a self-organizing artificial neural network based adaptive resonance theory. There is vast potential for the characteristics of its imitating the Human brain nerve system working in neurophysiology and psychology. But learning rate of ART2 can not be adjusted directly, model drift phenomenon of ART2 network occurs frequently. To solve this problem, this paper discusses the common-used learning rules of ART2 network at first and then it points out that although there is no learning rate in these learning rules as other artificial neural networks, but it implicitly exists. The way to adjust the learning rate is suggested and the suppression to pattern drift is verified by a vector learning trial. The categorization results to Iris dataset are also compared to illustrate the function of learning rate.