CLASSIFICATION OF S-AES ENCRYPTION KEY BITS USING MULTILAYER NEURAL NETWORKS
Abstract
Symmetric-key encryption plays a central role in contemporary cybersecurity, providing a fast and reliable means of protecting digital information. In this work, we explore whether the individual key bits of the Simplified Advanced Encryption Standard (S-AES) can be identified through machine learning, focusing on multilayer perceptron (MLP) neural networks. Using a dataset composed of plaintext–ciphertext pairs generated from randomly selected 16-bit keys, several neural models were trained under different hyperparameter configurations. The experiments reveal that some key bits are easier for the models to learn than others, indicating uneven sensitivity across the key space. These observations underscore the significance of proper hyperparameter tuning and point to potential applications in cryptanalysis studies.
References
[1] J. Daemen and V. Rijmen, The Design of Rijndael: AES — The Advanced Encryption Standard. Springer, 2002.
[2] NIST, “FIPS-197: Advanced Encryption Standard (AES),” National Institute of Standards and Technology, 2001.
[3] E. Schaefer, “A Simplified AES Algorithm for Educational Use,” Santa Clara University, 2003.
[4] W. Stallings, Cryptography and Network Security: Principles and Practice, 7th ed. Pearson, 2017.
[5] G. Benamira et al., “Neural Network-Based Approaches for Cryptanalysis: A Survey,” IEEE Access, vol. 8, pp. 145–162, 2020.
[6] S. Dubois and M. Robshaw, “Applying Machine Learning Techniques to Side-Channel Attacks,” in Proc. CHES, 2019.
[7] A. Hendrycks and K. Gimpel, “Gaussian Error Linear Units (GELUs),” arXiv:1606.08415, 2020.
