Function approximation using serial input neuron

An‐Chyau Huang

doi:10.1016/S0925-2312(01)00581-1

Function approximation using serial input neuron

2002-08-01

1

PRE

AI

DOI：10.1016/S0925-2312(01)00581-1

Original

Original request for help

Save

Abstract

En 中文

There is no general rule to specify proper complexity of conventional multilayer feedforward neural networks applied to function approximation applications where a specified approximation error is required. In this paper, we are going to suggest a neural computing structure based on a serial input neuron which can be used as an approximator of real valued functions define on the real line. The output of the approximator is proved to be exactly the value of a polynomial evaluated at the point represented by the given input of the approximator. The coefficients of the polynomial are obtained via learning. The order of the polynomial can be specified by the user and hence the approximation error and generalization can be controlled. Complexity of the proposed structure is independent to the unknown function to be approximated. It is the order of the approximating polynomial that determines computation complexity in terms of the number of iterations. (C) 2002 Elsevier Science B.V. All rights reserved.

Keywords:

artificial neural network

function approximation

Function approximation using serial input neuron

Abstract

Journal

Organization