A Joint Stochastic Gradient Algorithm and Its Application to System Identification with RBF Networks

Author

Chen, Badong ; Hu, Jinchun ; Li, Hongbo ; Sun, Zengqi

Author_Institution

State Key Lab. of Intelligent Technol. & Syst., Tsinghua Univ., Beijing

Volume

1

fYear

0

fDate

0-0 0

Firstpage

1754

Lastpage

1758

Abstract

Mean-square-error (MSE) and minimum-error-entropy (MEE) criteria play significant roles in adaptive filtering and learning theory. Nevertheless, both the criteria have their respective shortcomings. In this paper, we propose a more general and effective stochastic gradient algorithm under joint criterion of MSE and MEE, and derive the approximate upper bound for the step size in the adaptive linear neuron (ADALINE) training. In particular, we demonstrate the superiority of this joint adaptive algorithm by applying it into system identification with radial basis function (RBF) networks

Keywords

gradient methods; identification; learning (artificial intelligence); mean square error methods; minimum entropy methods; radial basis function networks; stochastic processes; RBF networks; adaptive linear neuron; joint stochastic gradient algorithm; mean-square-error; minimum-error-entropy; radial basis function networks; system identification; Entropy; Function approximation; Higher order statistics; Least squares approximation; Neurons; Radial basis function networks; Stochastic processes; Stochastic systems; System identification; Upper bound; ADA-LINE; MEE; MSE; RBF networks; Stochastic gradient algorithm;

fLanguage

English

Publisher

ieee

Conference_Titel

Intelligent Control and Automation, 2006. WCICA 2006. The Sixth World Congress on

Conference_Location

Dalian

Print_ISBN

1-4244-0332-4

Type

conf

DOI

10.1109/WCICA.2006.1712654

Filename

1712654