ISSN 1817-2172, рег. Эл. № ФС77-39410, ВАК

Differential Equations and Control Processes
(Differencialnie Uravnenia i Protsesy Upravlenia)

Randomized Unsupervized Algorithms for Neural Networks

Author(s):

O. Izmakova

Russia, 199034, St. Petersburg, Universitetskaya nab., 7/9,
St.Petersburg State University,
Department of Mathematics and Mechanics,

izmakova@inbox.ru

Abstract:

Two new unsupervised learning algorithms are described in this paper. They belong to the class of randomized stochastic algorithms.
The conditions for conver\-gence of estimates to the true values of unknown parameters are established for both algorithms.
The problem of the Hebb-Hopfield networks learning is discussed as an example. The method is given to solve the problem. It is based on the variational approach and randomized unsupervised algorithm.

Full text (pdf)