Registration Number
0223U001325,
0122U001731
, R & D reports
Title
Development of theoretical foundations and new methods of processing, recognition, forecasting of human-oriented processes and systems to solve problems of artificial intelligence
popup.stage_title
Head
Krak Yurii V., Доктор фізико-математичних наук
Registration Date
27-01-2023
Organization
Taras Shevchenko National University of Kyiv
popup.description2
Research object: methods and algorithms of machine learning, neurocomputing, forecasting, analysis of deterministic and stochastic models of human-oriented information, methods of classification and clustering the human-oriented data.
The purpose of the work: the development of theoretical foundations and new effective mathematical methods of researching and intelligent processing of large data sets of human-oriented processes and systems which are based on the approaches of artificial intelligence, machine learning, neural network technologies, forecasting, classification and clustering the information.
Research methods: algebraic methods, methods of qualitative theory of dynamic systems, statistical modeling of random processes, stochastic optimization algorithms, filtering algorithms, methods of information classification and clustering.
Using the Lyapunov direct method, the conditions of asymptotic stability and convergence of learning processes were established on continuous mathematical models of Hopfield neural networks. Significantly new results have been obtained for neural
networks with argument deviation, which are described in the terms of functional differential equations with a delay. Statistical models of random processes which are represented in the form of random series and stochastic integrals, as well as the models of random processes in the norms of different functional spaces have been studied.
Algorithms for finding the optimal path using stochastic evolutionary algorithms have been created. Statistical algorithms for correcting the baseline drift and filtering the electrocardiographic signals for effective diagnostic systems have been developed. The algorithms for improving the convolutional neural networks to investigate the humanoriented information have been created. The authors have been proposed the crossplatform technology for modeling and recognizing the gestures of the Ukrainian dactyl alphabet.
Product Description
popup.authors
Volchyna Iryna I.
Kasianiuk Veda S.
Manziuk Eduard A.
Pashko Anatolii O.
Stelya Оleg B.
Khusainov Denis Ya.
Shatyrko Andrii V.
Yaremenko Serhii O.
popup.nrat_date
2023-01-27
