Applications of hybrid learning to automated system design

The evolution of biological systems demonstrates the potential inherent in nonstructured performance-drive design processes for solving difficult design problems. A hybrid learning testbed is described that uses adaptive learning techniques which differ from conventional highly structured AI techniques and instead emulate nature's methods. The testbed incorporates genetic learning, neural networks, and clustering algorithms. The use of these techniques as a means of automating the design of pattern recognition systems is explored. The testbed provides a tangible focus for studying the key components of automated design: model representations, search strategies, and evaluation criteria. It demonstrates how a variety of adaptive techniques can be applied to the automated design of pattern recognition systems.<<ETX>>