Chaos-based Key Generator using Artificial Neural Networks Models

Abstract

Because of the growing number of attacks on existing systems, it is crucial to design new cryptographic techniques to ensure confidentiality. Although, the chaotic systems and deep learning methods have been proven to be effective in cryptography because of the nonlinearity and the high degree of randomness [1] and due to this, we combined them to produce cryptographic keys. This paper presents the chaotic time series forecasting models in order to produce encryption keys applied in cryptographic applications. Our artificial neural network (ANN) is trained by the Unified chaotic system samples using the optimal layers design. For comparative purposes, we have tested the performances of the Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) models in terms of chaotic time series patterns prediction. The Mean Squared Error (MSE) of the model achieves a value of 3.2x1$0^{-3}$ and we discover that, despite the small differences between the predicted samples of LSTM and GRU models, they produce nearly identical results for the task of chaotic time series prediction when compared to the MLP.