Cryptology ePrint Archive: Report 2022/183

Improving Differential-Neural Cryptanalysis with Inception Blocks

Liu Zhang and Zilong Wang and Boyang Wang

Abstract: In CRYPTO'19, Gohr proposed a new cryptanalysis strategy using machine learning algorithms. Combining the neural distinguisher with a differential path and integrating the advanced key recovery procedure, Gohr achieved a 12-round key recovery attack on Speck32/64. Chen and Yu improved accuracy of neural distinguisher considering derived features from multiple-ciphertext pairs instead of single-ciphertext pairs. Bao et al. presented the concept of generalized neutral bits, improved 12-round, and devised a practical 13-round key recovery attack on Speck32/64 by enhancing the differential path prepended on the top of neural distinguisher. To capture more dimensional information, inspired by the Inception Blocks in GoogLeNet, we use multiple parallel convolutional layers as the core of the network structure to train neural distinguisher. For Speck32/64, we improve the accuracy of (5-8)-round neural distinguisher and train a new 9-round neural distinguisher. For Simon32/64, we improve the accuracy of (7-11)-round neural distinguisher and train a new 12-round neural distinguisher. In addition, we extend the idea of neutral bits in Gohr's work to solve the requirement of the same distribution of multiple-plaintext pairs in key recovery attack. Under the combined effect of multiple improvements, the time complexity of our (11-13)-round key recovery attacks for Speck32/64 has been reduced to a certain extent. Surprisingly, the success rate of our 12-round key recovery attack can reach $100\%$. For Simon32/64, we increase the success rate and decrease the time complexity of 16-round key recovery attack. Also, we successfully implement a 17-round key recovery attack. Sadly, because the accuracy of 9-round neural distinguisher of Speck32/64 and 12-round neural distinguisher of Simon32/64 is lower, they can't be used for more rounds key recovery attack. The source codes are available in https://drive.google.com/drive/folders/17AWYupor_bX2rEe000tPuppdNH4dlmHm?usp=sharing.

Category / Keywords: Neural Distinguisher, Differential-Neural Cryptanalysis, Inception Blocks,Speck, Simon

Date: received 17 Feb 2022, last revised 12 Apr 2022

Contact author: 17lzhang3 at gmail com

Available format(s): PDF | BibTeX Citation

Version: 20220412:115759 (All versions of this report)

Short URL: ia.cr/2022/183


[ Cryptology ePrint archive ]