Radial Basic Networks M-training by Asymmetric Influence Functions

Abstract

We consider robust approach to radial basic networks training under the presence of measurement noise, which have asymmetric distributions. For minimization of the suggested asymmetric functionals the algorithms of Gauss−Newton and Levenberg−Marquardt are used. Estimation of noise parameters is done by the algorithm of stochastic approximation. Results of modeling, which confirm efficiency of the suggested approach, are stated.