Resilient backpropagation for RBF networks

Many algorithms have been proposed in order to train radial basis function (RBF) networks. Resilient Backpropagation (RPROP) with a weight decay term is proposed to train RBF networks and used to differentiate surfaces of 3D objects in range images and to classify eight different machine learning data sets for classification purposes. We show the advantages of resilient backpropagation for the RBF network structure within this classification context. The network structure is a combination of supervised and unsupervised learning layers. Experimental results show that a radial basis function network trained with resilient backpropagation can be successfully applied to differentiate surfaces of 3D objects in range images as well as to classification of machine learning problems.