Approximation and estimation techniques for neural networks

Abstract

The determination of the neural-path weights and other network parameters is posed as a state estimation problem. The application of the Kalman filter algorithm for training the neural network via this form of state estimation is suggested. Two cases of the problem are considered. The first one (the discrete case) is a linear estimation problem for the situation in which the given mapping (to be approximated) is specified in terms of a discrete, finite set of input-output pattern pairs. The second one (the continuous case) is a nonlinear estimation problem in which the given mapping is defined over a compact, non-discrete subset of R/sup n/.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>