TY - GEN
T1 - RLND
T2 - 24th International Conference on Applied Cryptography and Network Security, ACNS 2026
AU - Liu, Jie
AU - Xu, Junjie
AU - Qiu, Yunhao
AU - Liu, Qibo
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2027.
PY - 2027
Y1 - 2027
N2 - Block ciphers are widely used to ensure the security of data in the digital age, yet the techniques for analyzing them continue to advance. Recently, neural distinguishers have garnered significant attention due to their higher accuracy compared to traditional methods when analyzing block ciphers. In this work, we propose a novel neural distinguisher, named RLND, which can extract both spatial and temporal features. Specifically, we design a multi-scale convolutional kernel-based residual network (MSCK-ResNet) to fuse spatial features across different scales through varied receptive fields. To capture temporal dependencies, we construct a multi-layer LSTM (ML-LSTM) module with varying numbers of neurons. The optimal configuration of RLND, including the number of MSCK-ResNet blocks and ML-LSTM layers, is determined experimentally to maximize accuracy. We evaluate the performance of the proposed distinguisher on five lightweight block ciphers, demonstrating that it achieves up to 0.5% and 5.5% higher accuracy with single and multiple ciphertext pairs, respectively. This highlights the strong potential of RLND for practical key recovery.
AB - Block ciphers are widely used to ensure the security of data in the digital age, yet the techniques for analyzing them continue to advance. Recently, neural distinguishers have garnered significant attention due to their higher accuracy compared to traditional methods when analyzing block ciphers. In this work, we propose a novel neural distinguisher, named RLND, which can extract both spatial and temporal features. Specifically, we design a multi-scale convolutional kernel-based residual network (MSCK-ResNet) to fuse spatial features across different scales through varied receptive fields. To capture temporal dependencies, we construct a multi-layer LSTM (ML-LSTM) module with varying numbers of neurons. The optimal configuration of RLND, including the number of MSCK-ResNet blocks and ML-LSTM layers, is determined experimentally to maximize accuracy. We evaluate the performance of the proposed distinguisher on five lightweight block ciphers, demonstrating that it achieves up to 0.5% and 5.5% higher accuracy with single and multiple ciphertext pairs, respectively. This highlights the strong potential of RLND for practical key recovery.
KW - Differential analysis
KW - Key recovery
KW - Lightweight block cipher
KW - Neural distinguisher
UR - https://www.scopus.com/pages/publications/105047068690
U2 - 10.1007/978-3-032-32578-5_7
DO - 10.1007/978-3-032-32578-5_7
M3 - 会议稿件
AN - SCOPUS:105047068690
SN - 9783032325778
T3 - Lecture Notes in Computer Science
SP - 186
EP - 205
BT - Applied Cryptography and Network Security - 24th International Conference, ACNS 2026, Proceedings
A2 - Jee, Kangkook
A2 - Kate, Aniket
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 22 June 2026 through 25 June 2026
ER -