A Pseudo-Random Number Generator (PRNG) is any algorithm generating a sequence of numbers approximating properties of random numbers. These numbers are widely employed in mid-level cryptography and in software applications. Test suites are used to evaluate the quality of PRNGs by checking statistical properties of the generated sequences. These sequences are commonly represented bit by bit. This paper proposes a Reinforcement Learning (RL) approach to the task of generating PRNGs from scratch by learning a policy to solve a partially observable Markov Decision Process (MDP), where the full state is the period of the generated sequence, and the observation at each time-step is the last sequence of bits appended to such states. We use Long-Short Term Memory (LSTM) architecture to model the temporal relationship between observations at different time-steps by tasking the LSTM memory with the extraction of significant features of the hidden portion of the MDP’s states. We show that modeling a PRNG with a partially observable MDP and an LSTM architecture largely improves the results of the fully observable feedforward RL approach introduced in previous work.

Managing blood glucose levels in Type 1 Diabetes Mellitus (T1DM) is essential to prevent complications. Traditional insulin delivery methods often require significant patient involvement, limiting automation. Reinforcement Learning (RL)-based controllers offer a promising approach for improving automated insulin administration. We propose a Dual Proximal Policy Optimization (Dual PPO) controller for personalized insulin delivery in a hybrid closed-loop system. The controller optimizes patient-specific insulin bounds through a grid search on pre-trained models to manage both hyperglycemia and hypoglycemia. A safe-control mechanism prevents insulin administration when glucose levels drop below a predefined threshold. The system was evaluated on 10 in silico adult patients using the UVA/Padova simulator, with five-day randomized meal scenarios. The Dual PPO controller significantly improved Time in Range (TIR) (69.30% ±1.61) compared to a single PPO model (61.69% ±1.54). The system effectively reduced severe hyperglycemia while maintaining a low incidence of severe hypoglycemia. Unlike conventional open-loop methods such as Basal-Bolus (BBC) and Proportional-Integral-Derivative (PIDC) controllers, our system requires minimal patient interaction, eliminating the need for carbohydrate estimation. The Dual PPO controller enhances personalized insulin delivery in T1DM, improving glycemic control while reducing patient burden. This approach advances precision medicine in diabetes management, with potential for future real-world applications.