Efficient and Robust Keyed Hash Function Based on Artificial Neural Networks

In this paper, we propose a new dynamic key and message-dependent hash function based on Artificial Neural Networks (ANN) that satisfies the necessary security requirements with low computational complexity. It requires only two rounds of confusion and diffusion operations. Moreover, a dynamic non-invertible construction technique of a synaptic weight matrix is defined in order to ensure the one-way property. The proposed message authentication algorithm is evaluated in terms of desirable cryptographic properties. The results show that the proposed solution is robust due to the dynamic cryptographic primitives approach, which also results in reduced latency and required resources. The initial weight matrix is changed at regular time intervals depending on the application requirements. Recent enhancement in ANN hardware implementations enables the practical application of the proposed solution within wireless and mobile networks.