INFORMATION THEORY, LEARNING SYSTEMS, AND SELF-ORGANIZING CONTROL SYSTEMS
Abstract
Learning systems are defined on three levels of complexity, the trained system, the adaptive system, and the self-organizing system. The functional purpose of each is discussed in terms of the removal of noise and a theorem stating a necessary condition for adaption is stated and proven. The theorem is then applied to a simple form of adaptive system, and it is shown that for a linear threshold device employed within its 'natural capacity', two bits per weight are necessary. The information theoretic model of the self- organizing system is translated into a goal directed control system model. This self-organizing control system model is analyzed and shown to have a simple performance surface that can be searched by relaxation methods. Experiments are discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 31, 1967
- Accession Number
- AD0656082
Entities
People
- Malcolm Rucj Uffelman