Accession Number : ADA180607
Title : The Hopfield Model and Beyond.
Descriptive Note : Technical rept.,
Corporate Author : BROWN UNIV PROVIDENCE RI CENTER FOR NEURAL SCIENCE
Personal Author(s) : Bachmann, Charles M.
Report Date : 15 MAY 1987
Pagination or Media Count : 37
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.
Descriptors : *NERVE CELLS, *ARTIFICIAL INTELLIGENCE, *MATHEMATICAL MODELS, ALGORITHMS, ANALOG SYSTEMS, DIGITAL SYSTEMS, ERROR CORRECTION CODES, CAPACITY(QUANTITY), LAYERS, RESPONSE, MEMORY DEVICES, STORAGE, MODELS, NEUTRAL, SOLID STATE PHYSICS, INFORMATION RETRIEVAL, DATA STORAGE SYSTEMS, GLASS, SPINNING(MOTION).
Subject Categories : Operations Research
Distribution Statement : APPROVED FOR PUBLIC RELEASE