Simple and Optimal Greedy Online Contention Resolution Schemes

TL;DR

This paper introduces simple, optimal greedy online contention resolution schemes that improve upon previous methods for various combinatorial constraints, providing strong guarantees even against the most powerful adversaries.

Contribution

It presents the first simple, optimal greedy online contention resolution schemes for multiple constraints, including single-item, partition matroids, and transversal matroids, with proven optimality.

Findings

Achieved a 1/e approximation ratio for greedy OCRSs.

Improved upon previous greedy OCRS algorithms for key constraints.

Proved the optimality of the proposed greedy OCRSs in the single-item case.

Abstract

Real-world problems such as ad allocation and matching have been extensively studied under the lens of combinatorial optimization. In several applications, uncertainty in the input appears naturally and this has led to the study of online stochastic optimization models for such problems. For the offline case, these constrained combinatorial optimization problems have been extensively studied, and Contention Resolution Schemes (CRSs), introduced by Chekuri, Vondr\'{a}k, and Zenklusen, have emerged in recent years as a general framework to obtaining a solution. The idea behind a CRS is to first obtain a fractional solution to a (continuous) relaxation of the objective and then round the fractional solution to an integral one. When the order of rounding is controlled by an adversary, Online Contention Resolution Schemes (OCRSs) can be used instead, and have been successfully applied in settings such as prophet inequalities and stochastic probing. In this work, we focus on greedy OCRSs, which provide guarantees against the strongest possible adversary, an almighty adversary. Intuitively, a greedy OCRS has to make all its decisions before the online process starts. We present simple $1/e$ - selectable greedy OCRSs for the single-item setting, partition matroids and transversal matroids, which improve upon the previous state-of-the-art greedy OCRSs for these constraints. We also show that our greedy OCRSs are optimal, even for the simple single-item case.