Huai-Yang Zhong

Risk-Sensitive Reinforcement Learning with Smoothed Quantile Objectives

UCB-BQRL is developed, a model-based optimistic learning algorithm that maintains confidence sets for the transition kernel and plans using a lower-buffered quantile criterion, and an information-theoretic lower bound of $\Omega(H/\rho_\tau\sqrt{AT})$ for the regret of any algorithm dealing with a quantile objective fu...