bayesian-risk

This paper proposes a quantile Bayesian risk-aware MDP framework for online RL that adaptively balances robustness and exploration over time, providing theoretical regret bounds and demonstrating strong empirical performance.