摘要
本文提出了一类新的双向联想神经网络。与以往基于能量函数的神经网络不同,该网络通过在两个不同的Hilbert空间的交替投影进行联想,因而其动态特性和稳态解可以在信号空间上进行分析且具有快速自适应自学习等优点。该网络可做为异联想存贮器或分类器。本文给出了该网络在存贮和恢复二值模式时的实验结果,表明了其优良的性能。
A novel class of bidirectional associative neural networks (BANN) is proposed in this paper. In contrast to the more conventional technique of forming an energy metric for the neural network, the BANN performed by alternating projection between two different Hilbert space and so the BANN retain the better attributes, such as the ease of analysis in signal space, high accuracy of the steady-state solution and fast training. The network can be configured as either a heteroassociative memory or a classfier.Simulation results on storing and retrieving bipolar patterns in BANN's are presented to show its good performance.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1991年第3期23-29,共7页
Acta Electronica Sinica