A Bayesian interpretation for the Hopfield network

Author

Hancock, Edwin R. ; Kittler, Josef

Author_Institution

Dept. of Comput. Sci., York Univ., Heslington, UK

fYear

1993

fDate

1993

Firstpage

341

Abstract

The Hopfield network is a type of associative memory capable of recovering stored patterns when presented with their corrupted realizations. It is the intention of the authors to demonstrate that it has a Bayesian interpretation. The framework for their study is a relaxation method which involves direct probabilistic modeling of the pattern corruption mechanism. The parameter of this model is the memoryless probability of label errors on nodes of the network. It is demonstrated that the Hopfield model is a limit of the relaxation approach with precise physical meaning in terms of this parameter. This label-error probability allows understanding of more general properties of Bayesian pattern reconstruction by relaxation

Keywords

Bayes methods; Hopfield neural nets; probability; relaxation theory; Bayesian interpretation; Hopfield network; associative memory; direct probabilistic modeling; label-error probability; memoryless probability; pattern corruption mechanism; pattern reconstruction; relaxation method; Associative memory; Bayesian methods; Computer science; Computer vision; Cost function; Hamming distance; Hebbian theory; Labeling; Pattern recognition; Relaxation methods;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 1993., IEEE International Conference on

Conference_Location

San Francisco, CA

Print_ISBN

0-7803-0999-5

Type

conf

DOI

10.1109/ICNN.1993.298580

Filename

298580