Many online decision-making problems correspond to maximizing a sequence of submodular functions. In this work, we introduce sum-max functions, a subclass of monotone submodular functions capturing several interesting problems, including best-of-$K$-bandits, combinatorial bandits, and the bandit versions on $M$-medians and hitting sets. We show that all functions in this class satisfy a key property that we call pseudo-concavity. This allows us to prove $\big(1 - \frac{1}{e}\big)$-regret bounds for bandit feedback in the nonstochastic setting of the order of $\sqrt{MKT}$ (ignoring log factors), where $T$ is the time horizon and $M$ is a cardinality constraint. This bound, attained by a simple and efficient algorithm, significantly improves on the $\widetilde{\mathcal{O}}\big(T^{2/3}\big)$ regret bound for online monotone submodular maximization with bandit feedback. We also extend our results to a bandit version of the facility location problem.
Sum-max Submodular Bandits / S. Pasteris, A. Rumi, F. Vitale, N. Cesa Bianchi (PROCEEDINGS OF MACHINE LEARNING RESEARCH). - In: Proceedings of The 27th International Conference on Artificial Intelligence and Statistics / [a cura di] S. Dasgupta, S. Mandt, Y. Li. - [s.l] : ML Research Press, 2024. - pp. 2323-2331 (( Intervento presentato al 27. convegno International Conference on Artificial Intelligence and Statistics tenutosi a Valencia nel 2024.
Sum-max Submodular Bandits
A. Rumi;N. Cesa Bianchi
2024
Abstract
Many online decision-making problems correspond to maximizing a sequence of submodular functions. In this work, we introduce sum-max functions, a subclass of monotone submodular functions capturing several interesting problems, including best-of-$K$-bandits, combinatorial bandits, and the bandit versions on $M$-medians and hitting sets. We show that all functions in this class satisfy a key property that we call pseudo-concavity. This allows us to prove $\big(1 - \frac{1}{e}\big)$-regret bounds for bandit feedback in the nonstochastic setting of the order of $\sqrt{MKT}$ (ignoring log factors), where $T$ is the time horizon and $M$ is a cardinality constraint. This bound, attained by a simple and efficient algorithm, significantly improves on the $\widetilde{\mathcal{O}}\big(T^{2/3}\big)$ regret bound for online monotone submodular maximization with bandit feedback. We also extend our results to a bandit version of the facility location problem.| File | Dimensione | Formato | |
|---|---|---|---|
|
pasteris24a.pdf
accesso aperto
Tipologia:
Publisher's version/PDF
Dimensione
19.67 MB
Formato
Adobe PDF
|
19.67 MB | Adobe PDF | Visualizza/Apri |
|
SumMax-Compressed.pdf
accesso aperto
Tipologia:
Publisher's version/PDF
Dimensione
1.61 MB
Formato
Adobe PDF
|
1.61 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.




