On binary associative memories based on recurrent neural networks
Journal
Journal of the Chinese Institute of Engineers, Transactions of the Chinese Institute of Engineers,Series A/Chung-kuo Kung Ch'eng Hsuch K'an
Journal Volume
17
Journal Issue
1
Pages
55-62
Date Issued
1994
Author(s)
Abstract
Associative memory has been one of the focal points in recent'neural network research. In this paper, we propose a general model for binary associative memories based on a recurrent network structure. The proposed model is based on an evolution process that is similar to the political election process. The essence of the new model lies in the weighting junctions and how the system evolves according to the combined, weighted contribution from all stored patterns. With appropriate choice of weighting functions, new and efficient binary associative memories can be developed quite easily. The model therefore provides a solid foundation for the design of binary associative memories suitable for future electronic and optical technology. Furthermore, many well-known neural associative memories are shown to be special cases of this new model with appropriate reformulation of their respective evolution equation. The stability, hardware complexity, and storage capacity issues of these associative memories are also discussed. © 1993 Taylor & Francis Group, LLC.
SDGs
Type
journal article