Contextual Bandits with Packing and Covering Constraints: A Modular Lagrangian Approach via Regression
Контекстные бандиты с ограничениями на упаковку и покрытие: модульный лагранжевый подход на основе регрессии
2022-11-14
SCID: 54.1/k3udxtkr
Discuss with AI
Lagrangian approachcontextual bandits with linear constraintspacking and covering constraintsregression oraclesvanishing regret
Figures from the paper
Abstract (AI)
We consider contextual bandits with linear constraints (CBwLC), a variant of contextual bandits in which the algorithm consumes multiple resources subject to linear constraints on total consumption. This problem generalizes contextual bandits with knapsacks (CBwK), allowing for packing and covering constraints, as well as positive and negative resource consumption. We provide the first algorithm for CBwLC (or CBwK) that is based on regression oracles. The algorithm is simple, computationally efficient, and statistically optimal under mild assumptions. Further, we provide the first vanishing-regret guarantees for CBwLC (or CBwK) that extend beyond the stochastic environment. We side-step strong impossibility results from prior work by identifying a weaker (and, arguably, fairer) benchmark to compare against. Our algorithm builds on LagrangeBwK (Immorlica et al., FOCS 2019), a Lagrangian-based technique for CBwK, and SquareCB (Foster and Rakhlin, ICML 2020), a regression-based technique for contextual bandits. Our analysis leverages the inherent modularity of both techniques.
Key Findings
1
Avoids prior impossibility results by evaluating performance against a weaker and arguably fairer benchmark.
2
Combines the LagrangeBwK Lagrangian method with the SquareCB regression-based method through a modular analysis.
3
Introduces the first regression-oracle-based algorithm for contextual bandits with linear packing and covering constraints, including positive and negative resource consumption.
4
Provides the first vanishing-regret guarantees for constrained contextual bandits beyond purely stochastic environments.
5
The proposed algorithm is simple, computationally efficient, and statistically optimal under mild assumptions.
Research Object
Contextual bandits with linear packing and covering constraints, including positive and negative resource consumption
Research Subject
Regression-based Lagrangian algorithms, computational efficiency, statistical optimality, and vanishing-regret guarantees under stochastic and non-stochastic environments
Publication Details
Publication Date
2022-11-14
Journal
Publisher
ISSN
Cited by
2
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest