Parallel Deep Neural Network for Detecting Computer Attacks in Information Telecommunication Systems

The approach to parallelization of the deep neural network by dividing the training set into sub-set and training each sub-set into a separate copy of the model of the neural network, which allows to significantly reduce the training time and increase the reliability of the detection of attacks, is proposed. The structure of the neural network in the framework Caffe is developed, and experimental studies have been carried out that showed an increase in the reliability of the detection of attacks in comparison with known approaches.