An Immunological Approach to Initialize Centers of Radial Basis Function Neural Networks

The appropriate operation of a radial basis function (RBF) neural network depends mainly upon an adequate choice of the number and positions of its basis function centers. The simplest approach to train an RBF network is to assume fixed radial basis functions defining the activation of the hidden units, followed by the application of a regression procedure to determine the linear output weights. The main drawback of this strategy is the lack of an efficient algorithm to determine the amount and positions of the RBF centers. In this paper, an immunological approach makes use of the training data in order to initialize the radial basis functions. The approach to be proposed is inspired by the vertebrate immune system, and tries to compress the information contained in the data set while positioning the prototype vectors into representative regions of the input space. The algorithm is compared to random and k-means center selection, and results are reported concerning regression and classification problems.