Optimized deep neural network for cryptanalysis of DES

Abstract

Cryptography is the study of techniques which used to transforms the original text (plain text) to cipher text (non understandable text). Due to recent progress on digitized data exchange in electronic way, information security has become crucial in data storage and transmission. Some of the cryptographic algorithm has provided a promising solution which not only protects the data but also authenticates the systems and its participants, so the threat of various attacks is minimized. Nonetheless in the advancement of computing resources the cryptanalysis techniques also emerged and performing competitively in the field of information security with good results. In this paper, we have proposed the optimized deep neural network approach for cryptanalysis of symmetric encryption algorithm 64-bit DES (Data encryption standard). Our approach has used back propagation technique with multiple hidden layers and advanced activation function also we have addressed the problem of vanishing gradient. Further, the implementation results show that we have achieved 90% accuracy which is significantly higher as compared to previous approaches. We have also compared the proposed technique with the existing ones against three parameters i.e. time, loss, accuracy.