Composite squared-error algorithm for training feedforward neural networks

Author

Gonzaga, Dirceu ; De Campos, Marcello L R ; Netto, Sergio L.

Author_Institution

Dept. de Engenharia Electr., Inst. Mil. de Engenharia, Rio de Janeiro, Brazil

fYear

1998

fDate

5-6 Jun 1998

Firstpage

116

Lastpage

120

Abstract

A new algorithm, the so-called composite squared-error (CSE) algorithm, for training neural networks is presented. The CSE algorithm, whose roots lie in the field of adaptive IIR filtering, is able to avoid suboptimal solutions and associated saddle points, thus achieving lower values of the associated mean-squared-error function in a fewer number of iterations. For that matter, the CSE algorithm can regularly outperform other existing training schemes in most applications where neural networks are employed

Keywords

IIR filters; adaptive filters; adaptive signal processing; convergence of numerical methods; digital filters; error analysis; feedforward neural nets; filtering theory; learning (artificial intelligence); adaptive IIR filtering; backpropagation; composite squared-error algorithm; convergence; feedforward neural networks training; iterations; mean-squared-error function; Adaptive filters; Backpropagation algorithms; Convergence; Error correction; Feedforward neural networks; Filtering algorithms; Multi-layer neural network; Neural networks; Neurons; Signal processing algorithms;

fLanguage

English

Publisher

ieee

Conference_Titel

Advances in Digital Filtering and Signal Processing, 1998 IEEE Symposium on

Conference_Location

Victoria, BC

Print_ISBN

0-7803-4957-1

Type

conf

DOI

10.1109/ADFSP.1998.685707

Filename

685707