Oracle-Efficient Combinatorial Semi-Bandits

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kim, Jung-hun, Vojnović, Milan, Oh, Min-hwan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917039868739584
author Kim, Jung-hun
Vojnović, Milan
Oh, Min-hwan
author_facet Kim, Jung-hun
Vojnović, Milan
Oh, Min-hwan
contents We study the combinatorial semi-bandit problem where an agent selects a subset of base arms and receives individual feedback. While this generalizes the classical multi-armed bandit and has broad applicability, its scalability is limited by the high cost of combinatorial optimization, requiring oracle queries at every round. To tackle this, we propose oracle-efficient frameworks that significantly reduce oracle calls while maintaining tight regret guarantees. For the worst-case linear reward setting, our algorithms achieve $\tilde{O}(\sqrt{T})$ regret using only $O(\log\log T)$ oracle queries. We also propose covariance-adaptive algorithms that leverage noise structure for improved regret, and extend our approach to general (non-linear) rewards. Overall, our methods reduce oracle usage from linear to (doubly) logarithmic in time, with strong theoretical guarantees.
format Preprint
id arxiv_https___arxiv_org_abs_2510_21431
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Oracle-Efficient Combinatorial Semi-Bandits
Kim, Jung-hun
Vojnović, Milan
Oh, Min-hwan
Machine Learning
We study the combinatorial semi-bandit problem where an agent selects a subset of base arms and receives individual feedback. While this generalizes the classical multi-armed bandit and has broad applicability, its scalability is limited by the high cost of combinatorial optimization, requiring oracle queries at every round. To tackle this, we propose oracle-efficient frameworks that significantly reduce oracle calls while maintaining tight regret guarantees. For the worst-case linear reward setting, our algorithms achieve $\tilde{O}(\sqrt{T})$ regret using only $O(\log\log T)$ oracle queries. We also propose covariance-adaptive algorithms that leverage noise structure for improved regret, and extend our approach to general (non-linear) rewards. Overall, our methods reduce oracle usage from linear to (doubly) logarithmic in time, with strong theoretical guarantees.
title Oracle-Efficient Combinatorial Semi-Bandits
topic Machine Learning
url https://arxiv.org/abs/2510.21431