Search academic texts

Information × Registration Number 0213U003427, 0110U000458 , R & D reports Title Evolving hybrid systems of computational intelligence with variable structure for data mining popup.stage_title Head Bodyanskiy Yevgeniy Volodymyrovych, Доктор технічних наук Registration Date 05-02-2013 Organization Kharkiv National University of Radioelectronics popup.description2 The object of the study is intelligent data analysis in conditions of total or partial a priory and ongoing uncertainty and based on evolving hybrid systems of computational intelligence with variable structure. The purpose of the study is developing of a fuzzy-spiking-neural clustering network on analog-digital spike-neurons, able to work in conditions of variable number of clusters; research and developing of speed-optimal evolutionary algorithms and decision making processes in systems with interval membership functions and evolutionary algorithms of spatial-temporal analysis of video data; developing of methods, models and procedures for synthesis of neuro-fuzzy regulators with immune adjusting of the system's structure and parameters. The studies in area of 3rd generation of neural networks (spiking-neural networks based on a new paradigm of liquid computing) are conducted. These neural networks have some advantages over traditional networks, but both theoretically and practically there aren't known computational intelligence evolving hybrid systems based on this approach. That's why the study was focused on creation of new computational intelligence hybrid systems: self-learning spiking-neuro-fuzzy systems. Under this approach was developed an original hybrid self-learning fuzzy-spiking neural network for fuzzy clustering, which has improved performance in conditions of overlapping clusters in comparison with the known fuzzy clustering systems. For the first time was introduced fuzzy receptive neurons forming an input data fuzzification layer of spiking-neural networks that allow to improve the accuracy and processing speed in comparison with traditional population coding that are used in known spiking-neural network. For the first time was introduced and investigated analog-digital architectures of self-learning spiking-neural networks based on continuous and discrete Laplace transforms that allows to describe the functionality of liquid computing based neural networks' in terms of automatic control theory. It is experimentally proved that the number of training epochs of such hybrid systems is reduced to at least an order of magnitude in comparison with systems of second generation. Product Description popup.authors Іващенко Г.С. Аксак Н.Г. Бабенко В.О. Бабенко О.В. Балакірєва О.Г. Бесараб Д.А. Близнюк В.Г. Винокурова О.А. Власенко А.Н. Волкова В.В. Глушенкова І.С. Гончаренко М.О. Горпиненко Ю.С. Горшков Є.В. Гришко А.О. Гурін В.М. Дейнеко А. О. Долотов А.І. Дрюк А.Д. Задорожна Є.В. Кіріченко Л.О. Керносов М.А. Кирій В.В. Колчигін Б.В. Копаліані Д.С. Корабльов М.М. Костіна З.Л. Куценко Я.В. Кучеренко Є.І. Кушнарьов М.В. Лебьодкіна А.Ю. Міхнова О.Д. Макогон А.Е. Машошин Д.А. Машталір С.В. Мельнікова М.О. Мохаммад А.С. Плєхова Д.О. Плісс І.П. Попов С.В. Путятіна О.П. Радченко В.О. Рибальченко Т.В. Сліпченко О.В. Тімофєєв В.О. Танянський С.С. Творошенко І.С. Тесленко Н.О. Тищенко О.К. Ушков К.І. Філатов В.О. Фомічов О.О. Харченко О.О. Чапланова О.Б. Чепенко Т.Є. Чуб О.І. Шаламов М.О. Шафроненко А.Ю. Шкловець А.В. Шкуро К.О. popup.nrat_date 2020-04-02 Close