摘要
国外目前对人工神经网络的研究多数是从单个状态出发研究系统的输入输出关系或动态行为。本文有别于已有的方法,从系统的熵的角度出发对同步并行和异步串行计算进行分析,研究了神经网络系统的吸引性。本文论证了同步计算中神经网的熵取决于可能状态的总数而与状态分布概率无关;分析了能量函数与熵的关系,推导了Hopfield模型可能状态数目的上界。对于异步随机计算,本文分析了神经网系统的熵的收敛性以及它与状态迁移矩阵的特征值及迹的关系。
It is of interest to build up a unifying framework which explains the common nature of all the different Artificial Neural Network (ANN) models. The system-theoretical aspect of ANN is introduced from the standpoint of entropy. The convergent behavior of entropy in both synchronous and asynchronous neural computations is, analyzed, and, with respect to Hop-field CAM model, the upper bound of the number of possible states is derived. The relationship between entropy and state-transition matrix of ANN is also studied.
出处
《计算机学报》
EI
CSCD
北大核心
1990年第5期321-330,共10页
Chinese Journal of Computers
