NEURAL METHODS FOR CRYPTANALYSIS

Abstract

Artificial neural networks (ANN) are currently widely used for solving problems and tasks in various fields. In this term, cryptanalysis is no exception. New scientific approaches involving ANN in cryptographic systems studying and analyzing processes are proposed in the different stages of looking for weaknesses and breaking cryptographic algorithms. In this work, a short survey on scientific works related to the topic will be provided. Moreover, a specific ANN will be constructed for analyzing widely used asymmetric cryptosystem RSA. The model will be trained on generated training and testing data by encrypting random plaintexts. The work aims to achieve the highest performance and accuracy by trying various configurations like number of neurons, number of layers, activation functions and learning rate. Moreover, we will analyze how much data is required to train the network with good performance.