Research on Avalanche Effect of Neural Network Data Processing Algorithm

The avalanche effect of feedforward neural networks used for data encryption is studied in this paper. The selection regulation of the key elements of the neural network encryption algorithm is presented, including activation function, weight matrix, number of layers and bias vector, etc.; Then, the avalanche effect of this algorithm model is tested and analyzed. The results show that the encryption algorithm satisfied the SAC and high-order SAC ideally. When the plaintext changes 1-6 bit, the change bits of the ciphertext are concentrated on half of the length of the ciphertext, and complied with obviously normal distribution characteristics; the change bits of the ciphertext are concentrated in $\mathbf{2 5 \%} \sim \mathbf{7 5 \%}$ of the length of the ciphertext.