A Review of the Theory of Embedding Fields and Its Application to Neural Networks.

Abstract

The mathematical theory of embedding fields is reviewed from the point of view of its applicability to the study of neural networks. Emphasis is placed on those networks which can discriminate, learn, remember and reproduce complicated spatial-temporal patterns. Of particular interest is the realization of such networks using electronic models of neural elements termed syncoders. The theory of embedding fields allows an examination of networks composed of neuron-like elements and the demonstration that these elements can be arranged in a variety of ways to accomplish pattern discrimination, learning and remembering. Networks which are analyzed include those for learning both stationary and time-varying patterns as well as for pattern discrimination. Syncoders are determined to possess many of the attributes indicated by the theory to be required for the synthesis of networks which exhibit learning and memory capability. One characteristic which is found to be lacking and is a suggested area of future consideration, is the capability to establish an adequate dynamic associational strength at each cell. (Author)

Document Details

Document Type
Technical Report
Publication Date
Dec 01, 1972
Accession Number
AD0756919

Entities

People

  • Donald M. Levy

Organizations

  • University of Dayton Research Institute

Tags

DTIC Thesaurus Topics

  • Artificial Intelligence Computing
  • Artificial Intelligence Software
  • Demonstrations
  • Discrimination
  • Embedding
  • Learning
  • Neural Networks
  • Stationary

Fields of Study

  • Computer science

Readers

  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks
  • Microelectronics