APPLICATION OF NEURAL NETWORKS FOR THE DEVELOPMENT AND TESTING OF PSEUDORANDOM NUMBER GENERATORS

This article presents a review of recent research in the field of neurocryptography focused on pseudorandom number generators (PRNGs). Various types of PRNGs and their implementations are discussed, and the criteria for cryptographic security of generators are outlined. The reasons for selecting specific types of generators depending on the tasks are highlighted. The fundamentals of neural network (NN) theory are presented, and a comparison of their architectures in the context of PRNG design and output sequence testing is provided. Different sets of statistical tests for evaluating the quality of PRNG output sequences are reviewed. Key studies on PRNG generation based on neural networks, including classical recurrent networks (Elman, LSTM) and modern generative adversarial networks (GAN), are analyzed. Methods for testing PRNGs using neural networks are also discussed, emphasizing the potential negative consequences of underestimating this stage in the assessment of generator security.