English

High-Dimensional Sparse Linear Bandits

Machine Learning 2021-09-07 v2 Machine Learning Statistics Theory Statistics Theory

Abstract

Stochastic linear bandits with high-dimensional sparse features are a practical model for a variety of domains, including personalized medicine and online advertising. We derive a novel Ω(n2/3)\Omega(n^{2/3}) dimension-free minimax regret lower bound for sparse linear bandits in the data-poor regime where the horizon is smaller than the ambient dimension and where the feature vectors admit a well-conditioned exploration distribution. This is complemented by a nearly matching upper bound for an explore-then-commit algorithm showing that that Θ(n2/3)\Theta(n^{2/3}) is the optimal rate in the data-poor regime. The results complement existing bounds for the data-rich regime and provide another example where carefully balancing the trade-off between information and regret is necessary. Finally, we prove a dimension-free O(n)O(\sqrt{n}) regret upper bound under an additional assumption on the magnitude of the signal for relevant features.

Keywords

Cite

@article{arxiv.2011.04020,
  title  = {High-Dimensional Sparse Linear Bandits},
  author = {Botao Hao and Tor Lattimore and Mengdi Wang},
  journal= {arXiv preprint arXiv:2011.04020},
  year   = {2021}
}

Comments

Accepted by NeurIPS 2020

R2 v1 2026-06-23T19:59:36.633Z