Discovery and learning techniques for pattern recognition

Abstract

The problem of pattern recognition is that of grasping the meaning of complex entities. A great variety of problem areas attempt to cope with Just such complexities. The importance of discovery and induction methods embedded within a pattern recognition program lies in their potential generality. The program, rather than the programmer, can be asked to discover and learn about the problem area. Pattern recognition should ultimately provide powerful techniques for use in such research areas as form perception, target recognition, language translation, theorem proving, game playing, and the testing of psychological models. Research on some of these applications is already underway. To the extent that discovery and induction can be successfully employed by the program itself, they should be used. This paper describes a program that makes a beginning attempt at such use.