Neural network application to pseudorandom sequence generation simulation

The article discusses the methodology of using neural networks to simulate pseudorandom sequences, which allows finding the hidden structure and sequence algorithms to reduce the observed processes to deterministic ones. To improve the quality of simulation of sequences of generated numbers, models of recurrent neural networks are used, taking into account their ability to adapt to the generated continuous sequences. The paper proposes a method for using and tuning recurrent neural networks and the influence of selected hyperparameters that determine the internal structure, size of the input sequence, and length of the resulting generated sequence, which ultimately affects the quality and accuracy of the matching neural pseudorandom generator sequences and the real program equivalent. The obtained results show a possibility of using neural methods in the processes of hashing, predicting sequences of statistically random data, cryptographic algorithms and data compression algorithms, or finding the original process of generating the sequence under study.