Cognitive Networks for ATR: The Roles of Bifurcation and Chaos.

Abstract

In previous work we established the importance of developing neural networks that compute with diverse attractors as means for introducing cognition in neural networks. Cognition is the ability of a network to distinguish, on its own, between novel and familiar objects, the latter being the set of objects on which the network is trained and the former being the set of all objects never seen by the network before. Bifurcation between different kinds of attractors in such cognitive networks is the mechanism by which cognition is achieved. Cognition is essential in networks intended to operate in complex uncontrolled environment like that encountered in most automated target recognition (ATR) scenarios. For these reasons we began exploring neural net models that could exhibit all three types of attractors: point, periodic, and chaotic. There is mounting evidence from biology to indicate that the study of such networks can help understand higher-level brain functions such as feature-linking, cognition, separation of object from background, inferencing, and other higher-level functions and can lead to the development of higher-level neural networks with enhanced capabilities.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 19, 1994
Accession Number
ADA289010

Entities

People

  • N. H. Farhat

Organizations

  • University of Pennsylvania

Tags

Communities of Interest

  • Advanced Electronics

DTIC Thesaurus Topics

  • Cognition
  • Computer Simulations
  • Computers
  • Dynamics
  • Electrical Engineering
  • Engineering
  • Information Processing
  • Military Research
  • Neural Networks
  • Optical Storage
  • Recognition
  • Signal Processing
  • Simulations
  • Simulators
  • Spectra
  • Target Recognition

Fields of Study

  • Computer science

Readers

  • Artificial Intelligence
  • Control Systems Engineering.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks