THEORETICAL BASES OF THE METHOD OF POTENTIAL FUNCTIONS IN THE PROBLEM ABOUT INSTRUCTION OF AUTOMATIC MACHINES ABOUT SEPARATION OF ENTRANCE SITUATIONS OF CLASSES

Abstract : Algorithms of instruction of automatic machines for identification of classes of entrance situations are presented, on the basis of which lies the construction of socalled potential functions. The basic hypothesis for consideration is introduced about the character of functions, which divide sets corresponding to different classes of entrance situations. Proceeding from this hypothesis, theorems of convergence of algorithms for a finite number of steps are proved. Appears that these algorithms are realized by a broad class of diagrams. The characteristics of elements, from which are gathered the diagrams of this class, are practically arbitrary. It appears that the perceptron of Rosenblatt pertains to this class of diagrams, i.e., it is proved that the work of the perceptron may be comprehended as a realization of the method of potential functions. In connection with this the proven theorems of convergence of algorithms of potential functions resolve also the problem about convergence of the process in the perceptron. (Author)