Average-reward reinforcement learning studies sequential decision-making problems where the objective is to maximize the long-run average reward per time step. While discounted and finite-horizon RL have been extensively studied, both the theoretical understanding and practical development of average-reward RL remain relatively limited. Nevertheless, average-reward formulations are particularly important in practice for modeling continuing tasks without a natural endpoint, such as queueing systems, inventory control, and online platforms. In this project, students will learn the necessary background in average-reward RL, study its statistical properties from a theoretical perspective, and develop scalable algorithms for average-reward settings through numerical experiments.