English

Matroid Semi-Bandits in Sublinear Time

Machine Learning 2024-05-29 v1

Abstract

We study the matroid semi-bandits problem, where at each round the learner plays a subset of KK arms from a feasible set, and the goal is to maximize the expected cumulative linear rewards. Existing algorithms have per-round time complexity at least Ω(K)\Omega(K), which becomes expensive when KK is large. To address this computational issue, we propose FasterCUCB whose sampling rule takes time sublinear in KK for common classes of matroids: O(D polylog(K) polylog(T))O(D\text{ polylog}(K)\text{ polylog}(T)) for uniform matroids, partition matroids, and graphical matroids, and O(DK polylog(T))O(D\sqrt{K}\text{ polylog}(T)) for transversal matroids. Here, DD is the maximum number of elements in any feasible subset of arms, and TT is the horizon. Our technique is based on dynamic maintenance of an approximate maximum-weight basis over inner-product weights. Although the introduction of an approximate maximum-weight basis presents a challenge in regret analysis, we can still guarantee an upper bound on regret as tight as CUCB in the sense that it matches the gap-dependent lower bound by Kveton et al. (2014a) asymptotically.

Keywords

Cite

@article{arxiv.2405.17968,
  title  = {Matroid Semi-Bandits in Sublinear Time},
  author = {Ruo-Chun Tzeng and Naoto Ohsaka and Kaito Ariu},
  journal= {arXiv preprint arXiv:2405.17968},
  year   = {2024}
}
R2 v1 2026-06-28T16:43:30.671Z