DEEP NEURAL NETWORKS IN CYBER ATTACK DETECTION SYSTEMS

This article discusses potentials of mathematical support improvement of detection systems of remote cyber-attacks on network resources of data systems. Herewith, efficiency improvement of cyber-attacks management is achieved by models on the basis of deep neural networks. This is aided by appropriate neural network model which is pre-trained by means of sparse auto encoder. A deep neural network is trained by means of a set of algorithms simulating higher-level abstractions in analyzed data using architectures comprised of a set of non-linear transformations. The proposed model is supported by software which facilitated its approbation for detection of network cyber-attacks. The model testing demonstrated that the accuracy of its basic variant is comparable with that of modern detection systems of network cyber-attacks.