A fast robust learning algorithm for RBF network against outliers

Abstract

Training data set often contains outliers, which can cause substantial deterioration of the approximation realized by a neural network. In this paper, a fast robust learning algorithm against outliers for RBF network is presented. The algorithm uses the subtractive clustering(SC) method to select hidden node centers of RBF network, and the gradient descent method with the scaled robust loss function(SRLF) as the objective function to adjust hidden node widths and the connection weights of the network. Therefore, the learning of RBF network has robustness on dealing with outliers and fast rate of convergence. The experimental results show the advantages of the learning algorithm over traditional learning algorithms for RBF network. © Springer-Verlag Berlin Heidelberg 2006.