Hopfield网络中二元正交记忆模式的吸引域分析

在作为联想记忆的Hopfield网络中，二元正交记忆模式的分析对网络记忆容量的研究起着重要作用。文中提出了利用吸引指数的概念对各个二元正交记忆模式的吸引域进行估计的方法。理论分析和计算机仿真表明，当网络容量不超过0．33N时（比通常的0．15N要好），每个二元正交记忆模式的吸引域至少包含一个汉明球。

Analysis of Binary Orthogonal Memory Patterns in Hopfield Neural Networks

In associative Hopfield neural networks, analyzing binary orthogonal memory patterns plays a very important role in studying their storage capacity. The conception of attractive index is proposed to estimate the basin of attraction for each binary orthogonal memory patterns with components in Hopfield neural networks. Theoretic analysis and computer simulation show that each memory pattern's basin of attraction contains at least one Hamming sphere when the storage capacity is less than 0.33N which is better than usual 0.15N.