摘要
本文利用匹配滤波器组将Kosko提出的双向联想记忆(BAM) 建模为非线性闭环反馈系统,研究了模型中样本模式的稳定性与吸引性,给出并证明了稳定性和吸引性的几个充分条件。为了加强模型的记忆与误差校正能力,我们用指数非线性递增函数来提高检索模式与近邻样本的相关性,进一步讨论了修正后模型的稳定性与吸引性,分析表明模型的性能得到了很大改善。
Three problems are studied in the paper. The Bidirectional Associative Memory (BAM) is modeled by matched filters, so that a clearer idea of the information processing processes in the BAM model can be got. The stability and attractivity of exemplars in the BAM model are analysed in detail and several sufficient conditions for stability are given and proved. After be discussed the performance and limitation of the original BAM, a revised BAM model by taking an exponent function operating on the related correlations is presented. It is shown that the performances of the revised BAM model are significantly improved. Especially its memory capacity is nearly exponentially increased with min (N, P).
出处
《系统工程与电子技术》
EI
CSCD
1991年第6期9-15,共7页
Systems Engineering and Electronics
关键词
匹配滤波器
模型
记忆
神经网
Neural network, Associative memory, Matched filter.