A convergence analysis on a multilayered neural network using a discrete-time sigma-modified back propagation algorithm

摘要

In this paper, the learning gain of a discrete-time multilayered neural network is reconsidered from the viewpoint of the adaptive control systems. We present here a novel discrete-time learning law for a multilayered neural network, which is a sort of sigma-modified adaptive law used in the robust adaptive control systems. We also presents a brief proof of boundness of the estimator to be learned and a simple numerical simulation, where we show viability of the proposed learning law.