高阶神经网络模型的存储容量和吸引盆

高阶神经网络模型的存储容量和吸引盆 STORAGE CAPACITY AND ATTRACTION BASINS IN HIGHER-ORDER NEURAL NETWORK MODEL 王阿明 1 2 first-author 梁明理 1 2 武汉大学物理系 武汉大学物理系 武汉大学物理系 徐州医学院 武汉大学物理系 徐州医学院 本文研究了高阶神经网络模型的存储容量和存储图样的吸引盆,给出了理论推导和分析.对由30个神经元组成的二阶和三阶神经网络系统进行了计算机模拟计算.结果表明:（1）其存储容量分别为100和340个随机图样,远高于传统的Hopfield模型.（2）随着存储图样的数目增多,存储图样的吸引半径减少,因而其吸引盆减少.（3）高阶神经网络模型具有比Hopfield模型快得多的演化速率.高阶神经网络的高存储量和快演化速率可以克服Hopfield模型在应用方面的诸多不足,因此高阶神经网络模型有着良好的应用前景.本文还讨论了高阶神经网络模型的神经生物学基础. This paper inverstigates the storage capacity and attraction basins in higher-order neural network. The theoretical analyses are given. Calculaions of compute simulation to 2-order and 3-order neural network system with 30 neurons show that; （1 ）The storage capacity is approximately 100 and 390 patterns seperately, which is much higher than the capacity of traditional Hopfield’s model. （2） As the storage pattern number increases, its attration radius decreases so the attration basins decreases. （3） Higher-order neural network model is much faster in retrieval rate than Hopfield’s model. The high storage capacity and fast retreval rate in higher-order neural network model can overcome the weak- ness in Hopfield’s model, which enables higher-order neural network model to be of good prospects in application. The neurobiological basis of higher-order neural network model is discussed. 高阶神经网络 模型 存储容量 吸引半径 吸引盆 演化速率 higher-order neural network Model storage capacity attraction radius attraction basins retrieval rate Q61 1993-01-01 2021-04-01 1