A new fast multilayer perceptron training procedure based on the Davidon Fletcher Powell algorithm

Author

Abid, Sabeur ; Fnaiech, Furhut

Author_Institution

ESSTT, Tunis, Tunisia

Volume

1

fYear

2003

fDate

18-20 Sept. 2003

Firstpage

123

Abstract

The Davidon Fletcher Powell (DFP) optimization algorithm usually used for nonlinear least squares is presented and is combined with the standard backpropagation (SBP) algorithm yielding a new fast training multilayer perceptron (MLP) algorithm (DFP/SBP). The new algorithm is tested on several function approximation problems. The number of iterations required by this algorithm to converge is less than 40% of what is required by the SBP algorithm. Also it is less affected by the choice of initial weights and setup parameters. The DFP/SBP algorithm is much more efficient than either of other techniques when the network contains no more than few hundred weights.

Keywords

backpropagation; least squares approximations; multilayer perceptrons; optimisation; Davidon Fletcher Powell optimization algorithm; multilayer perceptron training; nonlinear least squares; standard backpropagation; Acceleration; Approximation algorithms; Backpropagation algorithms; Convergence; Least squares approximation; Least squares methods; Multilayer perceptrons; Newton method; Signal processing; Signal processing algorithms;

fLanguage

English

Publisher

ieee

Conference_Titel

Image and Signal Processing and Analysis, 2003. ISPA 2003. Proceedings of the 3rd International Symposium on

Print_ISBN

953-184-061-X

Type

conf

DOI

10.1109/ISPA.2003.1296880

Filename

1296880