ABSTRACT
Abstract: Y ong Y ao' sand Walter Freeman's KII and KIll models of the olfactory system (Yao and Free-
man, 1990; Yao, Freeman, Burke, and Yang, 1991; Freeman, 1992) describe neural network architectures based on the olfactory system that use chaotic dynamics to perform rapid featureclassification and novelty-detection. Engineering application of these KI, Kll, and KIll (,Katchalsky' 1) networks has been slowed by a number of issues: • Extension of the published results to practical problems has been difficult due to the limited
background of most computer engineers in the neurophysiology of olfaction. • Training methods for these networks remain poorly understood. • Lack of invariance in biological Katchalsky networks makes data encoding for engineering
applications unclear. This paper describes the lessons learned in investigating a KIll Katchalsky network for an aircraft radar application.
