Accession Number : AD0680102

Title :   CRITERION OF NUMBER OF RESOLVABLE ARGUMENTS FOR SELECTION OF USEFUL SIGNS FOR RECOGNITION AND PREDICTION SYSTEMS,

Corporate Author : FOREIGN TECHNOLOGY DIV WRIGHT-PATTERSON AFB OH

Personal Author(s) : Ivakhnenko, O. H.

Report Date : 10 MAY 1968

Pagination or Media Count : 25

Abstract : The author has developed a criterion of resolvable arguments for the selection of the most useful signs in recognition and prediction systems. The criterion makes it possible to find the minimum number of primary sensor units with which all images of the learning sequence can be divided with a multiple equal to unity. It is shown that the application of generalized signs (products of primary signs) does not change the selection of sensor units and can be used only to increase the multiplicity of resolution of the arguments to the value of q = 2/n where n is the number of sensor units. Thus, the selection of primary signs has as its aim finding the minimum set of signs with the multiple equal to unity. The selection of generalized signs has as its aim finding sets with any required multiplicity less than q = 2/n. Algorithms of searching are proposed which are analogous to the method of dynamic programming. These methods reduce considerably the volume of selection of probable combinations of signs for the selection of these two sets.

Descriptors :   (*LEARNING MACHINES, MATHEMATICAL PREDICTION), DYNAMIC PROGRAMMING, GAME THEORY, OPTIMIZATION, ALGORITHMS, USSR.

Subject Categories : Humanities and History
      Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE