摘要： Kosko的模糊联想记忆（FAM）神经网络得到了广泛的应用，然而却受到了低存储量的限制。Chung和Lee建立了一个矩阵存储多规则并能正确回忆的定理。基于这个定理，FAM模型的硬件及计算量需求能够显著地减少。但是，使用这个定理必须满足一个前提条件，即X(a)+X (a)(1，其中a=1, 2, ?, n。在这篇论文中，证明Chung和Lee的定理可以推广到连续的情形。提出了一个新的方法。这个方法即使在X(a)+X (a)> 1的情形下也是有效的，并且存储量可以进一步降低，更有利于大规则库系统的应用。

关键词: 模糊联想记忆, Sup-min垂直, 神经网络

Abstract: Kosko's fuzzy associative memory (FAM) suffers from low storage capacity. Chung and Lee established a theorem for perfect recall of all stored rules, based upon which the hardware and computation requirements of the FAM model can be reduced significantly. Unfortunately, X(a)+X (a)(1, where a=1, 2, ?, n, must be satisfied. In this paper, we first prove that Chung and Lee's theorem can be generalized to a continuous case. Then, we propose a new method which is effective even when X(a)+X (a)>1, where a=1, 2, ?, n, and whose storage can be further reduced because of the nature of the method.

中图分类号: 

• TP183