Search academic texts

Information × Registration Number 0213U001057, 0110U000015 , R & D reports Title Investigation of dynamic associative memory based on open recurrent neural networks. popup.stage_title Head Riznyk M. Olexandr, Registration Date 11-01-2013 Organization Institute of Mathematical Machines and System Problems NAS Ukraine. popup.description2 In the process, proposes a new concept of open dynamic system and the opportunity adaptive piecewise linear approximation of the dynamic behavior of associative memory, a structure similar to the nervous system of the cerebral cortex; the theory of dynamic neural associative memory created its programming model and experimentally investigated its ability to memorize and reproduce perturbed dynamic images in real time, an experimental model of distributed dynamic associative memory function that reproduces Hopfield bands cortex; proven capability self-healing associative memory contents in case of loss of neurons created its programming model, which first demonstrated the effect of treatment of amnesia when reminded of the vivid examples of the past, developed and tested a method for improving psevdorehulyaryzatsiyi training recurrent neural network application which reduces the error of multistep prediction by 10-30% , developed a new technology training recurrent neural networks for problems neyroupravlinnya that enhances the quality of training 5-10 times compared with the method of Kalman filter, and by 25-30% in comparison with the classical method of gradient descent. Product Description popup.authors А.О. Антоненко А.О. Куссуль Д.В. Новицький Д.О. Дзюба Д.О. Дзюба М.Е. Куссуль О.А. Калина О.Г. Садова О.М.Різник О.С. Сичов С.О. Мартьянов popup.nrat_date 2020-04-02 Close