Skip to main content
QUICK REVIEW

[论文解读] Scalable Auction Algorithms for Bipartite Maximum Matching Problems

Quanquan C. Liu, Yingyue Ke|arXiv (Cornell University)|Jan 1, 2023
Cryptography and Data Security被引用 1
一句话总结

本文提出了用于最大权二分匹配(MWM)和最大基数b-匹配(MCbM)的可扩展拍卖算法,在黑板模型中分别实现了 O(log n / ε⁸) 和 O(log n / ε²) 轮次。该工作解决了由 Dobzinski、Nisan 和 Oren(2014)提出的一个开放问题,即在通信复杂度较低的分布式、交互式环境中,能否通过拍卖算法实现 MWM 的 (1−ε)-近似解。与以往工作相比,该方法在半流式、工作-深度和大规模并行计算(MPC)模型中显著提升了空间和轮次效率。

ABSTRACT

Bipartite maximum matching and its variants are well-studied problems under various models of computation with the vast majority of approaches centering around various methods to find and eliminate augmenting paths. Beginning with the seminal papers of Demange, Gale and Sotomayor [DGS86] and Bertsekas [Ber81], bipartite maximum matching problems have also been studied in the context of auction algorithms. These algorithms model the maximum matching problem as an auction where one side of the bipartite graph consists of bidders and the other side consists of items; as such, these algorithms offer a very different approach to solving this problem that do not use classical methods. Dobzinski, Nisan and Oren [DNO14] demonstrated the utility of such algorithms in distributed, interactive settings by providing a simple and elegant O(log n/ε²) round maximum cardinality bipartite matching (MCM) algorithm that has small round and communication complexity and gives a (1-ε)-approximation for any (not necessarily constant) ε > 0. They leave as an open problem whether an auction algorithm, with similar guarantees, can be found for the maximum weighted bipartite matching (MWM) problem. Very recently, Assadi, Liu, and Tarjan [ALT21] extended the utility of auction algorithms for MCM into the semi-streaming and massively parallel computation (MPC) models, by cleverly using maximal matching as a subroutine, to give a new auction algorithm that uses O(1/ε²) rounds and achieves the state-of-the-art bipartite MCM results in the streaming and MPC settings. In this paper, we give new auction algorithms for maximum weighted bipartite matching (MWM) and maximum cardinality bipartite b-matching (MCbM). Our algorithms run in O(log n/ε⁸) and O(log n/ε²) rounds, respectively, in the distributed setting. We show that our MWM algorithm can be implemented in the distributed, interactive setting using O(log² n) and O(log n) bit messages, respectively, directly answering the open question posed by Demange, Gale and Sotomayor [DNO14]. Furthermore, we implement our algorithms in a variety of other models including the the semi-streaming model, the shared-memory work-depth model, and the massively parallel computation model. Our semi-streaming MWM algorithm uses O(1/ε⁸) passes in O(n log n ⋅ log(1/ε)) space and our MCbM algorithm runs in O(1/ε²) passes using O((∑_{i ∈ L} b_i + |R|) log(1/ε)) space (where parameters b_i represent the degree constraints on the b-matching and L and R represent the left and right side of the bipartite graph, respectively). Both of these algorithms improves exponentially the dependence on ε in the space complexity in the semi-streaming model against the best-known algorithms for these problems, in addition to improvements in round complexity for MCbM. Finally, our algorithms eliminate the large polylogarithmic dependence on n in depth and number of rounds in the work-depth and massively parallel computation models, respectively, improving on previous results which have large polylogarithmic dependence on n (and exponential dependence on ε in the MPC model).

研究动机与目标

  • 在分布式和流式模型中,为最大权二分匹配(MWM)和最大基数b-匹配(MCbM)开发可扩展的拍卖算法。
  • 解决 Dobzinski、Nisan 和 Oren(2014)提出的一个开放问题:交互式、低通信复杂度的拍卖算法是否能实现 MWM 的 (1−ε)-近似解。
  • 与现有最先进方法相比,降低半流式、工作-深度和大规模并行计算(MPC)模型中空间和轮次复杂度对 ε 的依赖。
  • 在工作-深度和 MPC 模型中,分别消除深度和轮次复杂度中对 n 的大体积多对数依赖。

提出的方法

  • 将 Demange、Gale 和 Sotomayor(1986)以及 Bertsekas(1981)提出的拍卖框架适配到加权匹配场景,采用修改后的价格更新规则。
  • 应用 Gupta-Peng 变换,将边划分为 O(log(1/ε)(W)) 个基于权重的桶和 O(1/ε) 份副本,从而通过并行或串行处理实现高效近似。
  • 使用黑板模型协调分布式计算,每个节点向中心协调器发送消息,以最小化总通信比特数。
  • 在多种模型中实现该算法:半流式模型(MWM 的轮次为 O(1/ε⁸),空间为 O(n log n · log(1/ε)))、共享内存工作-深度模型和 MPC 模型。
  • 利用拍卖算法的结构特性,控制通信和轮次复杂度,分别实现 MWM 和 MCbM 的 O(log n / ε⁸) 和 O(log n / ε²) 轮次。
  • 提出一种变换框架,可将任意 (1+ε)-近似 MWM 协议转化为 (1+16ε)-近似版本,且通信、工作或空间的膨胀程度在不同模型中可控。

实验结果

研究问题

  • RQ1拍卖算法是否能在通信复杂度较低的分布式、交互式环境中,实现 MWM 的 (1−ε)-近似解,如同其在最大基数匹配(MCM)中所实现的那样?
  • RQ2在半流式模型中,MWM 的轮次复杂度、通信成本和空间使用之间最优权衡为何?
  • RQ3拍卖方法如何被适配到具有单侧度数约束的 b-匹配问题?
  • RQ4拍卖算法能否在工作-深度和 MPC 模型中分别消除深度和轮次复杂度中对 n 的大体积多对数依赖?
  • RQ5在不同模型中,将一个 (1+ε)-近似 MWM 协议扩展为 (1+16ε)-近似版本所需的最小通信、工作或空间开销是多少?

主要发现

  • 所提出的 MWM 算法在黑板模型中运行时间为 O(log n / ε⁸) 轮,相比之前工作实现了显著改进,后者对 ε 具有指数依赖。
  • 该算法在分布式、交互式环境中实现了 MWM 的 (1−ε)-近似解,使用 O(log² n) 和 O(log n) 比特消息,直接回答了 Dobzinski、Nisan 和 Oren(2014)提出的开放问题。
  • 在半流式模型中,MWM 算法使用 O(1/ε⁸) 轮次和 O(n log n · log(1/ε)) 空间,对 ε 的空间依赖性相比最佳已知结果实现了指数级改进。
  • 对于 MCbM,该算法运行时间为 O(log n / ε²),空间为 O((∑ᵢ∈L bᵢ + |R|) log(1/ε)),相比先前方法,轮次复杂度更低,且对 ε 的空间依赖性更小。
  • 在工作-深度模型中,该算法消除了深度中对 n 的大体积多对数依赖,实现了 O(D(n, m, f(ε), ε) · log(1/ε)(W)) 的深度,且工作量膨胀可控。
  • 在 MPC 模型中,该算法减少了轮次复杂度中对 ε 的指数依赖,实现了每台机器 O(S(n, m, f(ε), ε) + n · log(1/ε)(W)) 的空间和总计 O(T(n,m,f(ε),ε)·log(1/ε)(W)/ε) 的总空间,优于先前结果。

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。