English

应用科学学报 ›› 1998, Vol. 16 ›› Issue (4): 397-402.

• 论文 • 上一篇    下一篇

一种基于RCAM结构的联想记忆器

王保云1, 何振亚2, 杨绿溪2   

  1. 1. 南京邮电学院;
    2. 东南大学
  • 收稿日期:1996-01-01 修回日期:1997-12-09 出版日期:1998-12-31 发布日期:1998-12-31
  • 作者简介:王保云:博士,讲师,南京邮电学院信息工程系,南京 210003
  • 基金资助:
    国家自然科学基金

A Novel Associative Memory Based on the Architecture of Recurrent Correlation Associative Memory

WANG BAOYUN1, HE ZHENYA2, YANG LÜ XI2   

  1. 1. Dept. of Inform. Eng., Nanjing Univ. of Post & Telecomm., Nanjing 210003;
    2. Dept. of Radio Eng., Southeast Univ., Nanjing 210096
  • Received:1996-01-01 Revised:1997-12-09 Online:1998-12-31 Published:1998-12-31

PDF

59

摘要： 递归相关联想记忆(RCAM)的回忆规则不同于Hopfield网络之处在于前者在输入与记忆模式的相关值上作用一非线性函数.在文献[7]的基础上,文中对所涉及的非线性函数进行了进一步的研究,提出了利用截断较小相关值来提高记忆性能的方法,得到了一种新的具有RCAM结构的联想记忆器(TRCAM).理论分析表明该方法可大大地提高记忆器对任意输入的信噪比,仿真实验也显示此方法可显著增大记忆模型在保证一定纠错能力下的记忆容量.

关键词: 联想记忆器, 神经网络, 容量, 纠错能力

Abstract: The main difference between the recall rule of recurrent correlation associative memory and that of Hopfield network is that RCAM operates a nonlinearity upon the correlations between input and stored patterns. In this paper, it proposes to improve the performance of RCAM by truncating the insignificant correlations, produce a novel associative memory based on the architecture of RCAM, noted as TRCAM. The theoretical analysis shows that the method raises the signal-to-noise ratio greatly. The simulations illustrate that this model enjoys a high capacity Back

Key words: associative memory, capacity, error-correcting capability, neural networks