Search academic texts

Information × Registration Number 0220U101292, 0118U003169 , R & D reports Title Methods of intellectual processing and analysis of large data based on deep neural networks popup.stage_title Head Sаchenko Anаtolii O., Доктор технічних наук Registration Date 07-02-2020 Organization Ternopil National Economic University popup.description2 Object of study - the processes of intelligent processing and analysis of big data using deep neural networks. The goal is to develop methods and tools for improving the efficiency and productivity of intelligent processing and analysis of big data based on deep neural networks. Research methods - methods of system analysis, mathematical statistics, data mining, big data, pattern recognition theory, artificial neural networks theory, neurocomputing, parallel computing. The results of the study and their novelty. Improved: the method of compression of big data on the example of network traffic parameters, which, unlike the known ones, is based on the use of deep autoencoder neural network to reduce the dimensionality of the analyzed information, allows to take into account the nonlinear nature of the neural elements and is characterized by minimizing the standard error of information recovery, existing methods and increase the reliability of the results obtained; the method of classification of big data on the example of attacks on information telecommunication networks using deep neural networks, which unlike known allows to analyze unstructured data without their preliminary processing, which allowed to increase the speed and made it possible to apply in intrusion detection systems in real time; neural network image recognition method based on image pre-processing that simplifies the localization of its individual parts and the subsequent recognition of localized blocks by means of a deep convolutional neural network and allows the recognition of low resolution images; deep neural network training method by splitting the training sample into sub-samples and parallel training of each sub-model on a separate copy of the neural network model, which increased the learning speed and reduced the use of GPU memory. Product Description popup.authors Bykovyy Pavlo Ye. Vasylkiv Nadiya M. Hladiy Hryhoriy M. Головко Володимир Адамович Koval Vasyl S. Komar Myroslav P. Кочан Володимир Володимирович Lipyanina-Honcharenko Khrysyna V. Lendiuk Taras V. Osolinskyy Oleksandr R. Roshchupkin Oleksiy Yu. Roshchupkina Nataliya V. Savenko Oleh S. Sachenko Anatoliy O. Shylinska Inna F. Yatskiv Vasyl V. popup.nrat_date 2020-04-02 Close