LINEAR SEPARABILITY OF SIGNAL SPACE AS A BASIS FOR ADAPTIVE MECHANISMS.
STANFORD RESEARCH INST MENLO PARK CALIF
Pagination or Media Count:
The report reviews the research results of the program entitled Linear Separability of Signal Space as a Basis for Adaptive Mechanisms. The major contributions of this program were two fold 1 the notion of discriminant functions was employed in constructing a framework for organizing past and present knowledge into a basis for further theoretical research on trainable pattern classifying machines, and 2 some significant new results were obtained on trainable pattern classifying machines. The specific research efforts reported fall into the following categories 1 investigation of the properties of various families of discriminant functions to be used by a pattern dichotomizer 2 investigation of various network structures for the implementation of useful families of discriminant functions and 3 investigation of various training rules to be used in selecting the appropriate discriminant function for a pattern dichotomizer. Author