一种扩展稳定吸引域的高效学习方法

An Efficient Learning Rule to Expand Stable Attraction Basins in Auto-Associative Neural Networks

基于Ｈｅｂｂ规划提出了一种能够稳定存储自联想模式向量，并基于各存储模式之间的汉明距离最大限度地扩展稳定模式吸引域的学习方法。采用本文方法对输入模式进行学习，各模式不但能够稳定存储，且具有一定的稳定吸引域，从而使网络具有较强的联想容错能力。

A repeated learning rule,which is based on Hebb rule the Hamming distance dis-tribution of the desired stable patterns,is studied in this paper.With the help of the proposedlearning rule,not only can the Iearned patterns be addressable,but an attraction basin withapredetermined radius is established for each attractor which enables the neural networkhave a strong auto-associative faulttolorance.