Adaptive Discretization in Online Reinforcement Learning

Adaptive Discretization in Reinforcement Learning Performance guarantees for RL algorithms are typically worst case instances, which pathological by design and not observed meaningful applications. Moreover, many domains (such as computer systems networking applications) have large state-action spaces require to execute with low latency. This phenomenon highlights a trifecta of goals practical algorithms: sample, storage, computational complexity. In this work, we develop an algorithmic framework nonparametric data-driven adaptive discretization. Our has provably better complexity than uniform discretization or kernel regression methods. highlight how the performance min-max optimal respect novel instance-specific measure that captures structure facility location newsvendor models.