摘要
提出了一种带有隐含神经元的单层神经网络模型。把网络的记忆容量区分为信息记忆容量和物理记忆容量。新模型能记忆相关图样,其信息容量α_i(最大记忆图样数/表达神经元数)首次超过了1。所作出的计算机模拟结果,表明了理论分析的正确性,证实了由5×5个显神经元组成的点阵能记住包含26个英文字母和4个任选图样的30个图样,因此该模型为神经网络的广泛应用提供了一条重要途径。
This paper presents a single layer neural network model with hidden neurons. This model can memorize correlated patterns and its information memory capacity α1 exceeds 1. we distinguish information memory capacity α1, from physical memory capacity αp in neural network. The results simulated by computer indicate that theory analysis in this paper is correct and the lattice consisted of 5 × 5 appearent neurons can memorize 30 patterns including 26 English letters and 4 patterns selected arbitrarily. Thus, this model gives an important way for application of neural network.
出处
《生物物理学报》
CAS
CSCD
北大核心
1992年第4期616-621,共6页
Acta Biophysica Sinica
关键词
神经网络
信息记忆容量
神经元模型
neural network, hidden neuron, information memory capacity,physical memory capacity
