RECURRENT NEURAL-NETWORK TRAINING WITH FEEDFORWARD COMPLEXITY

被引：24

作者：

OLUROTIMI, O ^{[1
]}

机构：

[1] GEORGE MASON UNIV,CTR C3I,FAIRFAX,VA 22030

来源：

IEEE TRANSACTIONS ON NEURAL NETWORKS | 1994年 / 5卷 / 02期

基金：

美国国家科学基金会;

关键词：

D O I：

10.1109/72.279184

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

This paper presents a training method that is of no more than feedforward complexity for fully recurrent networks. The method is not approximate, but rather depends on an exact transformation that reveals an embedded feedforward structure in every recurrent network. It turns out that given any unambiguous training data set, such as samples of the state variables and their derivatives, we need only to train this embedded feedforward structure. The necessary recurrent network parameters are then obtained by an inverse transformation that consists only of linear operators. As an example of modeling a representative nonlinear dynamical system, the method is applied to learn Bessel's differential equation, thereby generating Bessel functions within, as well as outside the training set.

引用

页码：185 / 197

页数：13

共 31 条

[1]

BANKS SP, 1988, MATH THEORIES NONLIN

[2]

Barhen J., 1990, ADV NEURAL INFORMATI, V2, P498

[3]

BARHEN J, APPLIED MATH LETT, V3, P13

[4] What Size Net Gives Valid Generalization? [J].