Cryptology ePrint Archive: Report 2021/311

Improved Neural Aided Statistical Attack for Cryptanalysis

Yi Chen and Hongbo Yu

Abstract: At CRYPTO 2019, Gohr improved attacks on Speck32/64 using deep learning. In 2020, Chen et al. proposed a neural aided statistical attack that is more generic. Chen etís attack is based on a statistical distinguisher that covers a prepended differential transition and a neural distinguisher. When the probability of the differential transition is pq, its impact on the data complexity is O(p^{-2}q^{-2}. In this paper, we propose an improved neural aided statistical attack based on a new concept named Homogeneous Set. Since partial random ciphertext pairs are filtered with the help of homogeneous sets, the differential transitionís impact on the data complexity is reduced to O(p^{−1}q^{−2}). As a demonstration, the improved neural aided statistical attack is applied to round-reduced Speck. And several better attacks are obtained.

Category / Keywords: secret-key cryptography / Cryptanalysis ∑ Deep learning ∑ Homogeneous set ∑ Statistical attack ∑ Speck families

Date: received 8 Mar 2021

Contact author: chenyi19 at mails tsinghua edu cn

Available format(s): PDF | BibTeX Citation

Version: 20210309:135130 (All versions of this report)

Short URL: ia.cr/2021/311


[ Cryptology ePrint archive ]