Deep learning combined with differential cryptanalysis has resulted in significant advancements in the field of cryptography. At the CRYPTO 2019 conference, Gohr introduced a novel approach by integrating deep learning with differential cryptanalysis to enhance the security of Speck32/64. The integration led to the development of a neural distinguisher that outperforms the existing DDT-based distinguisher. This breakthrough opens up opportunities for applying differential neural cryptanalysis techniques to other cryptographic algorithms.
A team of researchers, led by Liu Zhang, recently published their groundbreaking research in the journal Frontiers of Computer Science, co-published by Higher Education Press and Springer Nature. The team employed multiple convolutional layers with different kernel sizes, based on the round function of Simeck32/64, to capture the unique characteristics of the ciphertext in various dimensions. Compared to previous studies, their research demonstrated improved accuracy and increased rounds for the differential-neural distinguisher in Simeck32/64.
To achieve these results, the researchers enhanced the Inception neural network in alignment with the round function of Simeck32/64. By utilizing multiple ciphertext pairs as input to the neural network, they effectively captured the connections between them. These innovative approaches resulted in a significant improvement in the accuracy of the (9-12)-round differential-neural distinguisher (ND).
To establish a solid foundation for the ND, the research team conducted an extensive analysis of the differences induced by the input difference (0x0000, 0x0040) for Simeck32/64, computing the full distribution of differences for up to 13 rounds. In order to make a fair comparison with the ND, they also examined the accuracy of DDT-distinguishers (DD) using multiple ciphertext pairs under independent assumptions. The comparison clearly indicated that the 9- and 10-round NDs achieved higher accuracy than the DD, demonstrating that the ND contained more valuable information.
The researchers made significant progress by identifying simultaneous neutral bit-sets for a 3-round differential. Through comprehensive improvements, they were ultimately able to extend their findings to a practical 16 and 17-round key recovery attack for Simeck32/64 based on the ND.
The implications of this research are vast, as enhanced neural cryptanalysis methods have the potential to bolster the security of various cryptographic algorithms. This breakthrough opens up new possibilities in the field and highlights the importance of continued research in this area.
By merging deep learning with differential cryptanalysis, Gohr and his team have paved the way for enhanced security measures in cryptography. Their findings contribute to the ongoing advancements in the field and will undoubtedly shape future research and development. As the world becomes increasingly reliant on secure communication and data protection, the integration of artificial intelligence and cryptography will continue to play a critical role in safeguarding sensitive information.
Note: The DOI provided gives readers the opportunity to access the original research for further exploration and in-depth understanding of the topic.
