RESCALING OF VARIABLES IN BACK PROPAGATION LEARNING

被引：54

作者：

RIGLER, AK ^{[1
]}

IRVINE, JM ^{[1
]}

VOGL, TP ^{[1
]}

机构：

[1] ENVIRONM RES INST MICHIGAN,ANN ARBOR,MI 48107

来源：

NEURAL NETWORKS | 1991年 / 4卷 / 02期

关键词：

BACKWARD ERROR PROPAGATION; LAYERED NETWORKS; RESCALING; PRECONDITIONING;

D O I：

10.1016/0893-6080(91)90006-Q

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Use of the logistic derivative in backward error propagation suggests one source of ill-conditioning to be the decreasing multiplier in the computation of the elements of the gradient at each layer. A compensatory rescaling is suggested, based heuristically upon the expected value of the multiplier. Experimental results demonstrate an order of magnitude improvement in convergence.

引用

页码：225 / 229

页数：5

共 8 条

[1] ON A SUCCESSIVE TRANSFORMATION OF PROBABILITY-DISTRIBUTION AND ITS APPLICATION TO THE ANALYSIS OF THE OPTIMUM GRADIENT-METHOD [J].