A regularization term to avoid the saturation of the sigmoids in multilayer neural networks

被引：4

作者：

Garrido, L

Gomez, S

Gaitan, V

SerraRicart, M

机构：

[1] UNIV BARCELONA,FAC FIS,DEPT ESTRUCTURA & CONSTITUENTS MAT,E-08028 BARCELONA,SPAIN

[2] UNIV AUTONOMA BARCELONA,INST FIS ALTES ENERGIES,E-08193 BARCELONA,SPAIN

[3] INST ASTROFIS CANARIAS,E-38200 LA LAGUNA,SPAIN

来源：

INTERNATIONAL JOURNAL OF NEURAL SYSTEMS | 1996年 / 7卷 / 03期

关键词：

D O I：

10.1142/S0129065796000233

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

In this paper are 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.

引用

页码：257 / 262

页数：6

共 9 条

[1] COTTRELL GW, 1987, P 9 ANN C COGN SCI S, P462
[2] Hertz J., 1991, Introduction to the Theory of Neural Computation
[3] A PRACTICAL BAYESIAN FRAMEWORK FOR BACKPROPAGATION NETWORKS
MACKAY, DJC
[J]. NEURAL COMPUTATION, 1992, 4 (03) : 448 - 472
[4] OJA E, 1991, ARTIFICIAL NEURAL NETWORKS, VOLS 1 AND 2, P737
[5] PENNEBAKER W, 1990, JTC1SC2WG10JPEG8R8
[6] LEARNING REPRESENTATIONS BY BACK-PROPAGATING ERRORS
RUMELHART, DE
HINTON, GE
WILLIAMS, RJ
[J]. NATURE, 1986, 323 (6088) : 533 - 536
[7] OPTIMAL UNSUPERVISED LEARNING IN A SINGLE-LAYER LINEAR FEEDFORWARD NEURAL NETWORK
SANGER, TD
[J]. NEURAL NETWORKS, 1989, 2 (06) : 459 - 473
[8] DIMENSIONALITY-REDUCTION USING CONNECTIONIST NETWORKS
SAUND, E
[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1989, 11 (03) : 304 - 314
[9] MULTIDIMENSIONAL STATISTICAL-ANALYSIS USING ARTIFICIAL NEURAL NETWORKS - ASTRONOMICAL APPLICATIONS
SERRARICART, M
CALBET, X
GARRIDO, LS
GAITAN, V
[J]. ASTRONOMICAL JOURNAL, 1993, 106 (04) : 1685 - 1695

← 1 →