RBF networks training using a dual extended Kalman filter

A new supervised learning procedure for training RBF networks is proposed. It uses a pair of parallel running Kalman filters to sequentially update both the output weights and the centers of the network. The method offers advantages over the joint parameters vector approach in terms of memory requirements and training time. Simulation results for chaotic time series prediction and the 2-spirals classification problem are reported, and the effect of using 2 different pruning techniques for improving the generalization capacity is addressed. Index Terms – RBF networks, supervised training, Kalman filter, pruning