English

Instance-Wise Minimax-Optimal Algorithms for Logistic Bandits

Machine Learning 2021-03-10 v2 Machine Learning

Abstract

Logistic Bandits have recently attracted substantial attention, by providing an uncluttered yet challenging framework for understanding the impact of non-linearity in parametrized bandits. It was shown by Faury et al. (2020) that the learning-theoretic difficulties of Logistic Bandits can be embodied by a large (sometimes prohibitively) problem-dependent constant κ\kappa, characterizing the magnitude of the reward's non-linearity. In this paper we introduce a novel algorithm for which we provide a refined analysis. This allows for a better characterization of the effect of non-linearity and yields improved problem-dependent guarantees. In most favorable cases this leads to a regret upper-bound scaling as O~(dT/κ)\tilde{\mathcal{O}}(d\sqrt{T/\kappa}), which dramatically improves over the O~(dT+κ)\tilde{\mathcal{O}}(d\sqrt{T}+\kappa) state-of-the-art guarantees. We prove that this rate is minimax-optimal by deriving a Ω(dT/κ)\Omega(d\sqrt{T/\kappa}) problem-dependent lower-bound. Our analysis identifies two regimes (permanent and transitory) of the regret, which ultimately re-conciliates Faury et al. (2020) with the Bayesian approach of Dong et al. (2019). In contrast to previous works, we find that in the permanent regime non-linearity can dramatically ease the exploration-exploitation trade-off. While it also impacts the length of the transitory phase in a problem-dependent fashion, we show that this impact is mild in most reasonable configurations.

Keywords

Cite

@article{arxiv.2010.12642,
  title  = {Instance-Wise Minimax-Optimal Algorithms for Logistic Bandits},
  author = {Marc Abeille and Louis Faury and Clément Calauzènes},
  journal= {arXiv preprint arXiv:2010.12642},
  year   = {2021}
}

Comments

40 pages. AISTATS 2021, oral

R2 v1 2026-06-23T19:36:16.147Z