N.I. Chervyakov, A.A. Yevdokimov, E.V. Maslennikova

The structure of the multilayer neural network to build a cryptographic system with a public key. The system allows two users to securely share information for common key. Users then use the resulting open-th key to encrypt messages. The developed neural network cryptography system has an advantage over the use in practice of the Diffie-Hellman, which brings forth the calculation of pro-modulo a number of very high precision. The structure of the neural network is formed from a single layer of 256 neurons with sigmoid activation function. The number of inputs of neural networks is equal to 1, and the number of weights is 256. Multi-layer neural network is trained by adjusting the weights. Security system is based on the complexity of solving systems of nonlinear equations of the 256 with 257 real numbers, of which the private key can not be obtained. System does not require the use of large prime numbers for security. On the basis of the proposed neural network algorithm implemented encryption and digital signature.