S/N Analysis of Associative Memory Model based on Correlation Learning using a Non-Monotonic Property

• Kimoto Tomoyuki

  Oita National College of Technology

• Koizumi Kouji

  Faculty of Engineering Science, Osaka University

• Okada Masato

  Japan Science and Technology Corporation

Bibliographic Information

Other Title

• 非単調ニューロンを用いて相関学習された連想記憶モデルのＳ／Ｎ解析

Abstract

An associative memory model based on the correlation method using a non-monotonic neuron property is proposed and investigated by signal-to-noise analysis. The signal term for recalling and the cross-talk noise term preventing recalling are negatively correlated with each other in this model, so that the variance of the internal potential becomes much smaller than that of the cross-talk noise. This is contrasted with the situation of the Hopfield model in which the two terms are independent of each other. As a result, the present model has twice larger storage capacity in comparison with the Hopfiled model.

Journal

• The Brain & Neural Networks

  The Brain & Neural Networks 8 (3), 86-93, 2001

  Japanese Neural Network Society

Details

• CRID: 1390282679440497664
• NII Article ID: 10010424565
• NII Book ID: AA11658570
• DOI: 10.3902/jnns.8.86
• ISSN: 18830455
• ISSN: 1340766X
• Text Lang: ja
• Data Source:
  • JaLC
  • Crossref
  • CiNii Articles
• Abstract License Flag: Disallowed