English

Non-Asymptotic Sequential Tests for Overlapping Hypotheses and application to near optimal arm identification in bandit models

Statistics Theory 2021-11-19 v2 Statistics Theory

Abstract

In this paper, we study sequential testing problems with \emph{overlapping} hypotheses. We first focus on the simple problem of assessing if the mean μ\mu of a Gaussian distribution is smaller or larger than a fixed ϵ>0\epsilon>0; if μ(ϵ,ϵ)\mu\in(-\epsilon,\epsilon), both answers are considered to be correct. Then, we consider PAC-best arm identification in a bandit model: given KK probability distributions on R\mathbb{R} with means μ1,,μK\mu_1,\dots,\mu_K, we derive the asymptotic complexity of identifying, with risk at most δ\delta, an index I{1,,K}I\in\{1,\dots,K\} such that μImaxiμiϵ\mu_I\geq \max_i\mu_i -\epsilon. We provide non-asymptotic bounds on the error of a parallel General Likelihood Ratio Test, which can also be used for more general testing problems. We further propose lower bound on the number of observation needed to identify a correct hypothesis. Those lower bounds rely on information-theoretic arguments, and specifically on two versions of a change of measure lemma (a high-level form, and a low-level form) whose relative merits are discussed.

Keywords

Cite

@article{arxiv.1905.03495,
  title  = {Non-Asymptotic Sequential Tests for Overlapping Hypotheses and application to near optimal arm identification in bandit models},
  author = {Aurélien Garivier and Emilie Kaufmann},
  journal= {arXiv preprint arXiv:1905.03495},
  year   = {2021}
}