Computational methods in discrete mathematics

About using machine learning technologies for checking statistical properties of symmetric cryptography algorithms

A. A. Perov

Novosibirsk State University for Economics and Management

Abstract: This paper describes the use of machine learning technologies in cryptography, in particular, for carrying out the statistical analysis of block ciphers. The idea to adaptate ciphertexts to a model of neural network Inception V3 is stated. The author has developed a software utility for converting texts into JPEG. Conversion of ciphertexts for conducting experiments is completed. In the first experiment, the model distinguished absolutely all the ciphertexts of the Simon algorithm. The second experiment was to distinguish Simon cipher sequences on different rounds. The percentage of correct decisions on each subsequent round decreased. The total value approached 50 %. The third experiment showed an interesting scientific result, which consists in the ability to distinguish ciphertexts of different algorithms in the early rounds. In the fourth experiment, the model was trained on samples of early and full rounds. In 92 % of cases, the neural network made the right decisions to distinguish ciphertexts.

Keywords: cryptography, machine learning, statistical analysis, encryption round, iterative block cipher.

UDC: 519.7

DOI: 10.17223/2226308X/12/63

Bibliographic databases:

•