On the problem of synthesizing self-learning recognition systems
Alexander Fradkov
The problem of self-learning recognition systems The problem of self-learning pattern recognition is treated as the reconstruction
by an automaton of a certain apriori classification of the input objects. These are regarded as a sequence of points in Euclidean n-space forming a sample in an n-dimensional universe with an unknown continuous density distribution pi(x). Two algorithms that yield classification arbitrarily close to the prior one are described. The second algorithm correctly classifies only part of the input sample, but the size of the correctly classified part is proportional to the size of the sample. The results of computer experiments with both algorithms are described.
Published in Vestnik Leningrad Univ. Math., 1978 (5), pp.281-288.
(Translated from Вестник Ленинградского Университета,