Symmetric encryption with key distribution based on neural networks

Abstract

In this paper, a symmetric encryption approach with key distribution based on artifitial neural networks (ANN) has been investigated and implemented. We have used the ANN synchronization scheme with the tree parity machine for secret key exchange through a public channel. We developed the software for training the TPM and establishing the secure channel between two parties. We also made some experiments and got some conclusions about applying different TPM learning rules. We compare the number of tries and errors during TPM training. We found that using the classic Hebbian rule for TPM's training provides the best training results. We also found that more than 60% of TPM runs was learned from the weight changing. The developed software is suitable for application area which require reinforced protection of sending messages.