Implementation of a modified input output clustering algorithm for effective selection of RBF network centers

摘要

This study presents a new algorithm which extends an input-output clustering method for determining the centers of an RBF network. The proposed method uses the estimated lipschitz constant of a function as an initial weighting factor for augmenting training samples and apply a batch clustering method for determining augmented centers. Then, it adjusts this weighting factor by applying a gradient descent minimization based on the output error of RBF network. The simulations show that the proposed algorithm reduces the RBF error and provides a useful tool for center determination.