Accession Number : AD0696499

Title :   THE CONSTRUCTION OF A RECTIFYING SPACE FOR PATTERN RECOGNITION,

Corporate Author : FOREIGN TECHNOLOGY DIV WRIGHT-PATTERSON AFB OHIO

Personal Author(s) : Belyakov-Bodin,Vl. I.

Report Date : 12 MAR 1969

Pagination or Media Count : 15

Abstract : Automata termed 'artrons' (artificial neurons, i.e. neuron models or mathematical models of threshold elements), which have n binary inputs, a binary output and are capable of learning how to separate input situations into two classes, have recently begun to be often employed for pattern recognition. Two possible cases are considered. Either the reception field can be compared with coordinate unit vectors in Euclidean n-variate space X, such that the various input situations forming the set of nodes of the n-variate cube will be separated by a hyperplane in X (case 1), or the classes in X are not separated by a hyperplane and then transition from the X-space to the so-called 'rectifying' Z-space, in which separation of these classes by a hyperplane is feasible, is needed (case 2). The article examines certain aspects of the construction of the Z-space. The n-th degree polynomial is proposed as the logic separating function.

Descriptors :   (*PATTERN RECOGNITION, TRANSFORMATIONS(MATHEMATICS)), INFORMATION THEORY, MATHEMATICAL MODELS, AUTOMATA, POLYNOMIALS, MATHEMATICAL LOGIC, USSR

Subject Categories : Cybernetics
      Bionics

Distribution Statement : APPROVED FOR PUBLIC RELEASE