EVOLUTIONARY PATTERN RECOGNITION SYSTEMS.
ILLINOIS UNIV CHICAGO DEPT OF INFORMATION ENGINEERING
Pagination or Media Count:
An adaptive network is proposed which is capable of self-organization on two levels a fundamental level involving the structure of the system and a secondary level involving a set of variable weights within the structure. The system was simulated on a digital computer and its behavior investigated for the purpose of developing evolutionary processes by which systems could adapt structurally to their environment. The system was studied, first, as a non-evolutionary network adaptation could occur only through variation of the weights. An empirical equation was established relating the structural parameters of this system to its capabilities. The evolutionary system, possessing the capability of structure modification, was then investigated. Several heuristics were developed which allowed the system to significantly improve its performance in a pattern recognition task through structure modification. Author