Premium
An associative memory model derived from cross‐coupled Hopfield nets and its role in noise‐space dynamics
Author(s) -
Ozawa Seiichi,
Tsutsumi Kazuyoshi,
Baba Norio
Publication year - 1998
Publication title -
electrical engineering in japan
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.136
H-Index - 28
eISSN - 1520-6416
pISSN - 0424-7760
DOI - 10.1002/(sici)1520-6416(19981115)125:2<27::aid-eej4>3.0.co;2-q
Subject(s) - content addressable memory , hopfield network , bidirectional associative memory , artificial neural network , computer science , subspace topology , noise (video) , feed forward , autocorrelation , associative property , modular design , eigenvalues and eigenvectors , matrix (chemical analysis) , algorithm , topology (electrical circuits) , artificial intelligence , mathematics , pure mathematics , physics , engineering , statistics , materials science , composite material , quantum mechanics , control engineering , combinatorics , image (mathematics) , operating system
In this paper, the association characteristics of cross‐coupled Hopfield nets (CCHN) proposed as a modular neural network model are discussed analytically. In a CCHN, an arbitrary number of modules (Hopfield networks) can be mutually connected via feedforward networks called internetworks, whose output generates interactions among module networks. To evaluate the CCHN as a modular neural network, it has previously been applied to associative memory. Although its excellent association performance is supported by many simulation results, it is still difficult to compute the memory capacity exactly or to examine the dynamic properties rigorously, because CCHN information processing includes strong nonlinearity. Hence, as the first step to an analytical approach, this paper focuses on a single‐module CCHN whose interaction is realized by a two‐layered feedforward internetwork. In this case, the connection matrix of the CCHN degenerates into a single square‐matrix, as does a conventional auto‐association type of associative memory. Using eigenvalue analysis for the connection matrix, we reveal that the essential differences between the association characteristics of a CCHN and a conventional autocorrelation associative memory originate from dynamics in the noise‐space that is the orthogonal complement of the subspace generated from memory patterns. ©1998 Scripta Technica, Electr Eng Jpn, 125(2): 27–34, 1998