AVOIDING FALSE LOCAL MINIMA BY PROPER INITIALIZATION OF CONNECTIONS

被引：137

作者：

WESSELS, LFA ^{[1
]}

BARNARD, E ^{[1
]}

机构：

[1] UNIV PRETORIA,DEPT ELECTR & COMP ENGN,PRETORIA 0001,SOUTH AFRICA

来源：

IEEE TRANSACTIONS ON NEURAL NETWORKS | 1992年 / 3卷 / 06期

关键词：

D O I：

10.1109/72.165592

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

The training of neural-net classifiers is often hampered by the occurrence of local minima, which results in the attainment of inferior classification performance. It bas been shown [1] that the occurrence of local minima in the criterion function can often be related to specific patterns of defects in the classifier. In particular, three main causes for local minima were identified. Such an understanding of the physical correlates of local minima is important, since it suggests sensible ways of choosing the weights from which the training process is initiated. A new method of initialization is introduced which is shown to decrease the probability of local minima occurring on various test problems.

引用

页码：899 / 905

页数：7