A New Neural Distinguisher Considering Features Derived From Multiple Ciphertext Pairs

Abstract Neural-aided cryptanalysis is a challenging topic, in which the neural distinguisher ($\mathcal{ND}$) core module. In this paper, we propose new $\mathcal{ND}$ considering multiple ciphertext pairs simultaneously. Besides, are constructed from different keys. The motivation that distinguishing accuracy can be improved by exploiting features derived pairs. To verify motivation, have applied to five ciphers. Experiments show taking as input indeed brings improvement. Then, prove our applies two neural-aided key recovery attacks. Moreover, improvement helpful for reducing data complexity of statistic attack. code available at https://github.com/AI-Lab-Y/ND_mc.