Mutual Information and Machine Learning Based Distinguishers for Pseudo Random Bit Sequences

Abstract

In cryptology, a strong pseudo random bit sequence must be indistinguishable from pure random bit sequence or other strong pseudo random bit sequence. In this paper, distinguishers based on machine learning and mutual information are presented to distinguish various pseudo random bit sequences (generated by AES, 3DES, IDEA, Blowfish, RC4, and ‘rand’ function of C) from the strong pseudo random bit sequence of AES in CBC mode. Significant features are selected by using mutual information from the extracted features. SVM, Resilient Propagation ANN, and Multinomial Naive Bayes classifiers are used to distinguish. It is observed that pseudo random bit sequences of four block ciphers in ECB mode are distinguishable with more than 99% accuracy as compared to CBC and OFB mode. Bit sequences generated by RC4 stream cipher and ‘rand’ function of C are also significantly distinguishable.