THE SYNTHESIS OF MACHINES WHICH LEARN WITHOUT A TEACHER,
Abstract
Techniques of decision theory are applied to the problem of learning to recognize patterns without a teacher. As a result a generalized a posteriori probability computer is obtained which includes the solution of the problem of learning without a teacher, learning with a teacher, and no learning. The resulting equations are shown to describe a system which may be synthesized in delay feedback form, of fixed size, which is stable and converges to that system which would be optimum if a priori knowledge was available so that no learning was required. The solution is used to synthesize three systems in black box form: (1) a general system which learns to make binary decisions, a specific example of this system, and (3) a general system which learns to make multiple-category classifications. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1964
- Accession Number
- AD0443109
Entities
People
- S. C. Fralick
Organizations
- Stanford University