Regret Rates for $\epsilon$-Greedy Strategies for Nonparametric Bandits with Delayed Rewards

Sakshi Arya, Yuhong Yang

Keywords: Contextual Bandits, Delayed Rewards, Nonparametric Estimation, Nadaraya-Watson Estimator, Regret

TL;DR: Finite-time regret rates for Epsilon-Greedy in nonparametric Bandits with delayed rewards

Abstract: Incorporating delayed feedback is often crucial in applying multi-armed bandit algorithms in real-world sequential decision making problems. In this paper, we present finite time regret upper bounds for $\epsilon$-greedy type randomized allocation strategies in a nonparametric contextual bandits framework with delayed rewards. The strategies presented differ in how the exploration rate changes as a function of delays. We consider unbounded random delays and use the Nadaraya-Watson estimator for estimating the mean reward functions. We also propose practical data-driven strategies that adaptively choose between the two proposed strategies.

Primary Area: reinforcement learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8288