No-Regret Linear Bandits beyond Realizability
Abstract
We study linear bandits when the underlying reward function is not linear. Existing work relies on a uniform misspecification parameter $\epsilon$ that measures the sup-norm error of the best linear approximation. This results in an unavoidable linear regret whenever $\epsilon > 0$. We describe a more natural model of misspecification which only requires the approximation error at each input $x$ to be proportional to the suboptimality gap at $x$. It captures the intuition that, for optimization problems, near-optimal regions should matter more and we can tolerate larger approximation errors in suboptimal regions. Quite surprisingly, we show that the classical LinUCB algorithm -- designed for the realizable case -- is automatically robust against such gap-adjusted misspecification. It achieves a near-optimal $\sqrt{T}$ regret for problems that the best-known regret is almost linear in time horizon $T$. Technically, our proof relies on a novel self-bounding argument that bounds the part of the regret due to misspecification by the regret itself.
- Publication:
-
arXiv e-prints
- Pub Date:
- February 2023
- DOI:
- 10.48550/arXiv.2302.13252
- arXiv:
- arXiv:2302.13252
- Bibcode:
- 2023arXiv230213252L
- Keywords:
-
- Computer Science - Machine Learning
- E-Print:
- Proceedings of the 39th Conference on Uncertainty in Artificial Intelligence, PMLR 216:1294-1303, 2023