Learning laws exponential error convergence for recurrent neural networks

Abstract

In this paper, we propose new learning laws for adjusting the weights of recurrent high order neural networks (RHONN) when they are used to system identification problems. The main advantages of these learning laws over the classical robust adaptive ones, is that the identification error converges to zero exponentially fast, and that such a convergence is independent of the number of high order connections of the RHONN.