CONTINUOUS COMPLEX-VALUED BACKPROPAGATION LEARNING

被引：86

作者：

HIROSE, A

机构：

[1] Research Center for Advanced Science and Technology (RCAST), University of Tokyo, Meguro-ku, Tokyo 153

来源：

ELECTRONICS LETTERS | 1992年 / 28卷 / 20期

关键词：

NEURAL NETWORKS;

D O I：

10.1049/el:19921186

中图分类号：

TM [电工技术]; TN [电子技术、通信技术];

学科分类号：

0808 ; 0809 ;

摘要：

A novel back-propaption learning method is proposed for fully complex-valued layered neural networks. Nonlinearity suitable for realising smooth and unified complex learning is introduced. A gradient descent method is also analysed and optimised so that the variations of independent elements of the output complex vectors are related directly to the fragmentary changes of weighting matrix elements. The learning process is presented and demonstrated.

引用

页码：1854 / 1855

页数：2

共 7 条

[1] STATISTICAL NEURODYNAMICS OF ASSOCIATIVE MEMORY
AMARI, S
MAGINU, K
[J]. NEURAL NETWORKS, 1988, 1 (01) : 63 - 73
[2] STABILITY AND ATTRACTIVITY IN ASSOCIATIVE MEMORY NETWORKS
COTTRELL, M
[J]. BIOLOGICAL CYBERNETICS, 1988, 58 (02) : 129 - 139
[3] HIROSE A, 1992, IJCNN 92 BALTIMORE, V4, P152
[4] NEURAL NETWORKS AND PHYSICAL SYSTEMS WITH EMERGENT COLLECTIVE COMPUTATIONAL ABILITIES
HOPFIELD, JJ
[J]. PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA-BIOLOGICAL SCIENCES, 1982, 79 (08): : 2554 - 2558
[5] Kohonen T., 1988, SELF ORG ASS MEMORY
[6] KONO T, 1990, IEICE TECH REP, P7
[7] TSURUOKA H, 1992, INNS 91, P39

← 1 →