摘要
本文从Hopfield模型出发,提出了一个新的学习律。它能够在临界饱和状态下完成新图样的学习,同时保留了遗忘态的信息。讨论了该学习律与传统的Hebb学习律的区别,对其相应的联想演化行为给予理论上的分析。对400个神经元、存储图样数分别为30、40、50的系统给出了计算机模拟计算结果。
This paper presents a new learning rule according to Hopfield model. The rule can learn new pattern near full storage state and retain some information of forgetting pattern simultaneously. The diferentiation betweeen this rule and the generalized Hebb rule has been discussed. The behavior of associative evolution has been analysed theoretically. Calculations of computer simulation to systems With 400 neurons and 30, 40, 50 random patterns separately are given and analysed.
出处
《武汉大学学报(自然科学版)》
CSCD
1992年第3期31-36,共6页
Journal of Wuhan University(Natural Science Edition)
关键词
神经网络
饱和态
学习模式
neural netowork
saturation state
associative memory
correlation
synaptic strength