ArticlesNo Access

A REGULARIZATION TERM TO AVOID THE SATURATION OF THE SIGMOIDS IN MULTILAYER NEURAL NETWORKS

Departament d’Estructura i Constituents de la Matèria, Facultat de Física, Universitat de Barcelona, Diagonal 647, E-08028 Barcelona, Spain

Institut de Física d’Altes Energies, Universitat Autònoma de Barcelona, E-08193 Bellaterra (Barcelona), Spain

Search for more papers by this author

SERGIO GÓMEZ

Departament d’Estructura i Constituents de la Matèria, Facultat de Física, Universitat de Barcelona, Diagonal 647, E-08028 Barcelona, Spain

Present address: Dept. d’Enginyeria Informàtica (ETSE), Univ. Rovira i Virgili, Crt. de Salou s/n (Complex educatiu), E-43006 Tarragona (Spain).

Search for more papers by this author

VICENS GAITÁN

Institut de Física d’Altes Energies, Universitat Autònoma de Barcelona, E-08193 Bellaterra (Barcelona), Spain

Search for more papers by this author

, and

MIQUEL SERRA-RICART

Instituto de Astrofísica de Canarias, E-38200 La Laguna (Tenerife), Spain

Search for more papers by this author

https://doi.org/10.1142/S0129065796000233Cited by:3 (Source: Crossref)

Abstract

In this paper we propose a new method to prevent the saturation of any set of hidden units of a multilayer neural network. This method is implemented by adding a regularization term to the standard quadratic error function, which is based on a repulsive action between pairs of patterns.

This research is partly supported by the ‘Comissionat per Universitats i Recerca de la Generalitat de Catalunya’ and by EU under contract number CHRX-CT92-0004.