Extreme learning machine based sub-key generation for cryptography system

Abstract

The key generation process is the substantial step in any cryptosystem. Incorporating Artificial Neural Network (ANN) in the algorithmic work of cryptography achieves good performance in realizing high accuracy and security. In this paper, ANN based sub-key generation algorithm is presented. Extreme learning Machine (ELM) type is adopted for one hidden layer neural network. Initial key includes all needed information about ANN topology, activation function, and seeds for Pseudo-Random Number Generation (PRNG) in each round to initialize input-hidden layer weights and data. Sub-key in each round is generated from output layer weights. Evaluation measures have proved complete sensitivity and inevitability of this approach. In addition, it contributes in reducing the risks of breaking the symmetric key algorithms due to the generated independent sub-key in each round. Thus, it can be integrated in any cryptosystem for subkey generation.