Accession Number : ADA189315

Title :   Associative Memory Biological and Mathematical Aspects.

Descriptive Note : Technical rept.,

Corporate Author : MASSACHUSETTS INST OF TECH LEXINGTON LINCOLN LAB

Personal Author(s) : Eggers, Mitchell

PDF Url : ADA189315

Report Date : 29 Dec 1987

Pagination or Media Count : 50

Abstract : A tutorial is presented encompassing both biological and mathematical aspects of associative memory for pattern processing. A systems viewpoint is adopted whereby biological associative memory is viewed as a system of adaptive filters, with the free parameters of the filter corresponding to the strengths of the biological neural connections. Certainly such viewpoint is not intended to accurately depict the true mechanisms underlying the extraordinary capabilities of biological associative memory-fast pattern recognition and apparently infinite memory capacity. For such mechanisms will unlikely be discovered in the absence of tools allowing the observance of collective behavior over systems of neurons. However, the viewpoint does serve to integrate both mathematics and biology on a general level. Of most significance is perhaps the systematic treatment of mathematical associative memory. In the adaptive filter framework, associative memory is described and compared to traditional statistical techniques. Also, new insight into the generalization capability of associative memory is expressed. Conditions are presented to ensure both correct memory recall and significant generalization. Keywords: Optical processing; Artificial intelligence.

Descriptors :   *ASSOCIATIVE PROCESSING, *MEMORY DEVICES, *OPTICAL PROCESSING, ADAPTIVE FILTERS, BEHAVIOR, BIOLOGY, CAPACITY(QUANTITY), NERVE CELLS, PATTERNS, PROCESSING, RECALL, STATISTICAL PROCESSES, ARTIFICIAL INTELLIGENCE, NEURAL NETS, NETWORKS, PATTERN RECOGNITION, MATHEMATICAL MODELS, PARALLEL PROCESSING, COMPUTER ARCHITECTURE

Subject Categories : Computer Hardware
      Cybernetics
      Photography

Distribution Statement : APPROVED FOR PUBLIC RELEASE