Online Budget Allocation with Censored Semi-Bandit Feedback

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Bachoc, François, Cesa-Bianchi, Nicolò, Cesari, Tommaso, Colomboni, Roberto
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911421528276992
author Bachoc, François
Cesa-Bianchi, Nicolò
Cesari, Tommaso
Colomboni, Roberto
author_facet Bachoc, François
Cesa-Bianchi, Nicolò
Cesari, Tommaso
Colomboni, Roberto
contents We study a stochastic budget-allocation problem over $K$ tasks. At each round $t$, the learner chooses an allocation $X_t \in Δ_K$. Task $k$ succeeds with probability $F_k(X_{t,k})$, where $F_1,\dots,F_K$ are nondecreasing budget-to-success curves, and upon success yields a random reward with unknown mean $μ_k$. The learner observes which tasks succeed, and observes a task's reward only upon success (censored semi-bandit feedback). This model captures, for instance, splitting payments across crowdsourcing workers or distributing bids across simultaneous auctions, and subsumes stochastic multi-armed bandits and semi-bandits. We design an optimism-based algorithm that operates under censored semi-bandit feedback. Our main result shows that in diminishing-returns regimes, the regret of this algorithm scales polylogarithmically with the horizon $T$ without any ad hoc tuning. For general nondecreasing curves, we prove that the same algorithm (with the same tuning) achieves a worst-case regret upper bound of $\tilde O(K\sqrt{T})$. Finally, we establish a matching worst-case regret lower bound of $Ω(K\sqrt{T})$ that holds even for full-feedback algorithms, highlighting the intrinsic hardness of our problem outside diminishing returns.
format Preprint
id arxiv_https___arxiv_org_abs_2508_05844
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Online Budget Allocation with Censored Semi-Bandit Feedback
Bachoc, François
Cesa-Bianchi, Nicolò
Cesari, Tommaso
Colomboni, Roberto
Computer Science and Game Theory
Machine Learning
We study a stochastic budget-allocation problem over $K$ tasks. At each round $t$, the learner chooses an allocation $X_t \in Δ_K$. Task $k$ succeeds with probability $F_k(X_{t,k})$, where $F_1,\dots,F_K$ are nondecreasing budget-to-success curves, and upon success yields a random reward with unknown mean $μ_k$. The learner observes which tasks succeed, and observes a task's reward only upon success (censored semi-bandit feedback). This model captures, for instance, splitting payments across crowdsourcing workers or distributing bids across simultaneous auctions, and subsumes stochastic multi-armed bandits and semi-bandits. We design an optimism-based algorithm that operates under censored semi-bandit feedback. Our main result shows that in diminishing-returns regimes, the regret of this algorithm scales polylogarithmically with the horizon $T$ without any ad hoc tuning. For general nondecreasing curves, we prove that the same algorithm (with the same tuning) achieves a worst-case regret upper bound of $\tilde O(K\sqrt{T})$. Finally, we establish a matching worst-case regret lower bound of $Ω(K\sqrt{T})$ that holds even for full-feedback algorithms, highlighting the intrinsic hardness of our problem outside diminishing returns.
title Online Budget Allocation with Censored Semi-Bandit Feedback
topic Computer Science and Game Theory
Machine Learning
url https://arxiv.org/abs/2508.05844