摘要
关于Ｈｏｐｆｉｅｌｄ网的状态转移轨迹、吸引子、吸引域和如何逃离局部极小点等问题，尤其是对具有一般权系数和偏置的网络，一直没有得到很好的解决．本文首先定义了Ｈｏｐｆｉｅｌｄ网的关联网络，分析了关联网络的性质，找出了具有关联关系的网络状态转移轨迹之间存在的关联关系和规律．从而，如果已知了一个网络的状态转移轨迹，利用这一规律可以很容易地得出与其相关联的网络的状态转移轨迹，并给出了它们的吸引子之间关系的结论，为Ｈｏｐｆｉｅｌｄ网络的局部极小点的逃离、最不容错网络的判断提供了方法，为对网络的进一步分析和设计提供了一种新的方法和途径．所进行的大量计算机仿真实验验证了这一规律的存在．
The correlated network of a Hopfield network is defined and the properties of correlated networks are analyzed.The relationship between the state transferences of any two correlated networks on the correlating law is found out,with which the state transference of a Hopfield network can be easily deduced by that of any network correlated with it.The relationship between the attractors of any two correlated networks is derived out,which is helpful for the Hopield network to escape from the local minimum and for the judgement of network s tolerant ability to input data.Finally,computer simulation results are presented to verify the existance of the correlating law.
出处
《自动化学报》
EI
CSCD
北大核心
1997年第4期446-454,共9页
Acta Automatica Sinica
基金
国家"八六三"高科技资助
北京大学视觉与听觉信息处理国家重点实验室资助
关键词
HOPFIELD网络
关联分析
吸引子
人工神经网络
Hopfield network,correlating analysis,attractor,state transference,tolerant ability of a network.
