Accession Number : AD0676211

Title :   HIERARCHICAL DESCRIPTIONS, UNIVERSAL SPACES AND ADAPTIVE SYSTEMS.

Descriptive Note : Technical rept.,

Corporate Author : MICHIGAN UNIV ANN ARBOR LOGIC OF COMPUTERS GROUP

Personal Author(s) : Holland, John H.

Report Date : AUG 1968

Pagination or Media Count : 58

Abstract : The power of an adaptive system depends critically upon its ability to exploit common factors in successful techniques. Each time a device is tried, information accrues about components of each of the potential decompositions. Thus, the richer the variety of decompositions, the higher the effective sampling rate. Of course, to exploit this information about components, the adaptive system must use it to infer the performance of untried devices. And these inferences must, in turn, be used to plan which devices should be generated and tried next. At each stage, the flexibility and success of the process depends upon the flexibility and richness of the system's analysis and synthesis procedures. Among the many important procedures are those of: Substitution, Abstraction, Refinement, Modeling, Change of Representation, and Metacontrol. A structural formalism well-attuned to such procedures will exhibit three critical characteristics: Hierarchical Description, Self-Applicability, and Incorporation of Models. Formalisms with these characteristics are studied using a broad class of automaton representations, the class of compositions, which includes countably infinite devices such as von Neumann's cellular space and iterative circuit computers. (Author)

Descriptors :   (*ARTIFICIAL INTELLIGENCE, FEASIBILITY STUDIES), (*LEARNING MACHINES, THEORY), MATHEMATICAL LOGIC, ADAPTIVE SYSTEMS, AUTOMATA, ALGEBRA.

Subject Categories : Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE