Welcome to the resource topic for 2025/1306
Title:
Rethinking Learning-based Symmetric Cryptanalysis: a Theoretical Perspective
Authors: Yufei Yuan, Haiyi Xu, Lei Zhang, Wenling Wu
Abstract:In this paper, we revisit the standard approach to constructing neural distinguishers in symmetric cryptanalysis and introduce a game-like model, the Coin-Tossing model, to generalize this methodology. From the perspective of Boolean functions, we show that many classical cryptanalytic techniques can be generalized as a specific family of Boolean functions, termed the CPF class. We further explore the connection between learning CPF Boolean functions in the Coin-Tossing model and the well-known Learning Parity with Noise (LPN) problem. Leveraging the theoretical analysis, we identify key attributes of CPF functions that significantly affect how effectively machine learning algorithms can learn them. To validate our conclusions, we also conduct extensive experiments based on machine learning algorithms. Incorporating our theoretical insights, we propose an advanced 8-round and 9-round neural distinguisher for SPECK32/64 by reducing the problem complexity. Additionally, we propose a method based on the Goldreich-Levin algorithm to analyze and interpret what black-box distinguishers learn. Using this approach, we reinterpret several established neural distinguishers in terms of Fourier expansion. It is able to resolve the previous neural distinguisher in several Fourier terms. Notably, we identify a new type of distinguisher from neural networks that has not been discovered by cryptanalysts, which can be considered as a variant of the Differential-Linear distinguisher. We also demonstrate that the neural network not only learned the optimal Differential-Linear (DL) distinguishers found using existing MILP/MIQCP models, but also discovered even superior ones.
ePrint: https://eprint.iacr.org/2025/1306
See all topics related to this paper.
Feel free to post resources that are related to this paper below.
Example resources include: implementations, explanation materials, talks, slides, links to previous discussions on other websites.
For more information, see the rules for Resource Topics .