Using Machine Learning Technologies for Carrying out Statistical Analysis of Block Ciphers

This article presents the application of machine learning technologies to cryptography tasks, in particular, the statistical analysis of block ciphers. The author uses the Inception V3 neural network model, which is traditionally used for images recognition. A technology for adapting ciphertexts to the developed technique is proposed. The results of experiments on encrypted sequences are presented. The advantages of statistical analysis by machine learning over the classical methods of statistical testing are revealed. In this article, we demonstrate that in some cases using the machine-learning techniques it is possible to distinguish reduced-round ciphers ciphertexts from the random sequences having a smaller samples comparing with traditional statistical methods.