Horizon-Free and Variance-Dependent Reinforcement Learning for Latent Markov Decision Processes

Runlong Zhou, Ruosong Wang, Simon S. Du

ICML 2023

We provide an algorithm framework for Latent MDPs (with context in hindsight), achieving the first horizon-free minimax regret. We complement the study by giving a novel regret lower bound for LMDPs using the symmetrization technique.

Download here