Symmetric-key encryption remains a cornerstone of modern cryptographic security, offering an efficient mechanism for securing digital communication. This study investigates the feasibility of classifying key bits of the Simplified Advanced Encryption Standard (S-AES) using machine learning techniques, particularly multilayer perceptron (MLP) neural networks. A dataset of plaintext–ciphertext pairs generated from random 16-bit encryption keys is used to train multiple neural models with varying hyperparameters. The results demonstrate that certain key bits exhibit higher learnability than others, suggesting non-uniform model sensitivity across the key space. The findings emphasize the importance of hyperparameter selection and highlight potential implications for cryptanalysis research.

Symmetric encryption, S-AES, key bit classification, neural networks

J. Daemen and V. Rijmen, The Design of Rijndael: AES — The Advanced Encryption Standard. Springer, 2002.

NIST, “FIPS-197: Advanced Encryption Standard (AES),” National Institute of Standards and Technology, 2001.

E. Schaefer, “A Simplified AES Algorithm for Educational Use,” Santa Clara University, 2003.

W. Stallings, Cryptography and Network Security: Principles and Practice, 7th ed. Pearson, 2017.

G. Benamira et al., “Neural Network-Based Approaches for Cryptanalysis: A Survey,” IEEE Access, vol. 8, pp. 145–162, 2020.

S. Dubois and M. Robshaw, “Applying Machine Learning Techniques to Side-Channel Attacks,” in Proc. CHES, 2019.

A. Hendrycks and K. Gimpel, “Gaussian Error Linear Units (GELUs),” arXiv:1606.08415, 2020.