Cryptanalysis of SM4 algorithm based on convolutional residual networks

Due to the natural similarity between cryptanalysis and deep learning, various deep learning technologies have been applied to cryptanalysis. In order to analyze the security of state secret SM4 algorithm, the convolution residual network is used to build a model and search the difference divider of SM4 algorithm. The model is trained based on the selected plaintext difference and data set. Through data processing, parameter and function optimization, a 3-8 round difference divider is constructed. The model can distinguish the ciphertext pairs encrypted by SM4 algorithm with low number of rounds from random data, but with the increase of the number of rounds, the model can not effectively distinguish the ciphertext pair from the random data. Results show that SM4 algorithm has good security.