41
30

Improved Variance-Aware Confidence Sets for Linear Bandits and Linear Mixture MDP

Abstract

This paper presents new \emph{variance-aware} confidence sets for linear bandits and linear mixture Markov Decision Processes (MDPs). With the new confidence sets, we obtain the follow regret bounds: For linear bandits, we obtain an O~(poly(d)1+k=1Kσk2)\tilde{O}(poly(d)\sqrt{1 + \sum_{k=1}^{K}\sigma_k^2}) data-dependent regret bound, where dd is the feature dimension, KK is the number of rounds, and σk2\sigma_k^2 is the \emph{unknown} variance of the reward at the kk-th round. This is the first regret bound that only scales with the variance and the dimension but \emph{no explicit polynomial dependency on KK}. When variances are small, this bound can be significantly smaller than the Θ~(dK)\tilde{\Theta}\left(d\sqrt{K}\right) worst-case regret bound. For linear mixture MDPs, we obtain an O~(poly(d,logH)K)\tilde{O}(poly(d, \log H)\sqrt{K}) regret bound, where dd is the number of base models, KK is the number of episodes, and HH is the planning horizon. This is the first regret bound that only scales \emph{logarithmically} with HH in the reinforcement learning with linear function approximation setting, thus \emph{exponentially improving} existing results, and resolving an open problem in \citep{zhou2020nearly}. We develop three technical ideas that may be of independent interest: 1) applications of the peeling technique to both the input norm and the variance magnitude, 2) a recursion-based estimator for the variance, and 3) a new convex potential lemma that generalizes the seminal elliptical potential lemma.

View on arXiv
Comments on this paper