Fast and efficient sequential learning algorithms using direct-link RBF networks

Abstract

Novel fast and efficient sequential learning algorithms are proposed for direct-link radial basis function (DRBF) networks. The dynamic DRBF network is trained using the recently proposed decomposed/parallel recursive Levenberg Marquardt (PRLM) algorithm by neglecting the interneuron weight interactions. The resulting sequential learning approach enables weights to be updated in an efficient parallel manner and facilitates a minimal update extension for real-time applications. Simulation results for two benchmark problems show the feasibility of the new training algorithms.