English

An Improved Relaxation for Oracle-Efficient Adversarial Contextual Bandits

Machine Learning 2023-11-13 v2 Data Structures and Algorithms Machine Learning

Abstract

We present an oracle-efficient relaxation for the adversarial contextual bandits problem, where the contexts are sequentially drawn i.i.d from a known distribution and the cost sequence is chosen by an online adversary. Our algorithm has a regret bound of O(T23(Klog(Π))13)O(T^{\frac{2}{3}}(K\log(|\Pi|))^{\frac{1}{3}}) and makes at most O(K)O(K) calls per round to an offline optimization oracle, where KK denotes the number of actions, TT denotes the number of rounds and Π\Pi denotes the set of policies. This is the first result to improve the prior best bound of O((TK)23(log(Π))13)O((TK)^{\frac{2}{3}}(\log(|\Pi|))^{\frac{1}{3}}) as obtained by Syrgkanis et al. at NeurIPS 2016, and the first to match the original bound of Langford and Zhang at NeurIPS 2007 which was obtained for the stochastic case.

Keywords

Cite

@article{arxiv.2310.19025,
  title  = {An Improved Relaxation for Oracle-Efficient Adversarial Contextual Bandits},
  author = {Kiarash Banihashem and MohammadTaghi Hajiaghayi and Suho Shin and Max Springer},
  journal= {arXiv preprint arXiv:2310.19025},
  year   = {2023}
}

Comments

Appears in NeurIPS 2023

R2 v1 2026-06-28T13:05:06.754Z