摘要
本文提出了一种具有局域连接特性的二维神经网络模型,有效地克服了Hopfield种经网络因完全连接而带来的若干缺陷,并给出了相应的学习算法及用此神经网络记忆和识别26个英文字符的模拟结果.
In this paper, a new two dimensional neural network is firstly presented, in which each neuron is connected to the neurons, in its nearest neighbour local area. The network with local interconnection overcomes some drawbacks of Hopfield network, which is fully connected. Also in this paper, a new learning algorithm is put forth to permit the recall of general (non-orthogonal) patterns. The network was modeled on a computer and successfully recognized a set of non-orthogonal patterns.
出处
《计算机学报》
EI
CSCD
北大核心
1992年第7期541-545,共5页
Chinese Journal of Computers
基金
国家自然科学基金
关键词
神经网络
局域连接
学习算法
Two dimensional neural networks, local interconnection, learning algorithm.
