Formation of pseudorandom sequences based on neural network generative models with an attention mechanism

The algorithmic generationproblem of pseudorandom sequences is considered. To improve them, it is proposed to use a combination of a generative-predictive approach (a modification of generative-adversarial neural networks that show good results in generating images) and an attention mechanism that is used in large language models for generating text. According to the conducted experiments, this approach allowed to improve the proportion of sequences that passed the randomness test by 3.5 %, and the average probability of observing test statistics by 16 %. Keywords artificial neural networks, generative-predictive networks, neural networks-transformers, attention mechanism, generation of pseudo-random sequences