The Hopfield Model and Beyond.

Abstract

The standard Hopfield model (both digital analog) and algorithms to improve its performance are reviewed. An analysis of the model and the modification algorithms is given. Future directions for continuous models which have both large capacity and good error-correcting capabilities are examined. In 1982, Hopfield proposed a neutral model of memory storage and retrieval based on the theory of spin glasses in solid state physics. In the model, neurons are binary-valued threshold units, taking either the value 0 or 1 in one version, or 1 or -1 in an alternative version. This digital restriction of the neurons represents the neuron in two possible states-a 1 represents a neuron that is firing, while a 0 or a -1, a neuron that is inactive. Mathematically, this corresponds to replacing the experimentally observed neuronal input-output relationship, a graded response which can be characterized by a sigmoid function, with a step-function. The neurons form a single layer and are completely interconnected, with the strength of these connections, or synapses, given by a correlation matrix formed from the memory states to be stored in the system.

Document Details

Document Type
Technical Report
Publication Date
May 15, 1987
Accession Number
ADA180607

Entities

People

  • Charles M. Bachmann

Organizations

  • Brown University

Tags

DTIC Thesaurus Topics

  • Algorithms
  • Mathematics
  • Physics
  • Solid State Physics
  • Standards
  • Step Functions

Readers

  • Auditory Neuroscience/Auditory Physiology.
  • Computer Programming and Software Development.
  • Theoretical Analysis.