Results from the communication complexity literature have demonstrated that stable matching requires communication: one cannot find or verify a stable match without having access to essentially all of the ordinal preference information held privately by the agents in the market. Stated differently, these results show that stable matching mechanisms are not robust to even a small number of labeled inaccuracies in the input preferences. In practice, these results indicate that agents must go through the time-intensive process of accurately ranking each and every potential match candidate if they wish for the resulting match to be guaranteedly stable. Thus, in large markets, communication requirements for stable matching may be impractically high.
A natural question to ask, given this result, is whether some higher-order structure in the market can indicate which large markets have steeper communication requirements. In this paper, we perform such an analysis in a regime where agents have a utility-based notion of preference. We consider a dynamic model where agents only have access to an approximation of their utility that satisfies a universal multiplicative error bound. We apply guarantees from the theoretical computer science literature on low-distortion embeddings of finite metric spaces to understand the communication requirements of stable matching in large markets in terms of their structural properties. Our results show that for a broad family of markets, the error bound may not grow faster than $n^2\log(n)$ while maintaining a deterministic guarantee on the behavior of stable matching mechanisms in the limit. We also show that a stronger probabilistic guarantee may be made so long as the bound grows at most logarithmically in the underlying topological complexity of the market.
Queen’s University Belfast(贝尔法斯特女王大学);Freie Universität Berlin and Berlin School of Economics(柏林自由大学和柏林经济学院);Instituto de Matemática Aplicada San Luis, Universidad Nacional de San Luis, and CONICET(圣路易斯应用数学研究所、圣路易斯国立大学和阿根廷国家科学研究委员会);University of California, Berkeley(加州大学伯克利分校)
Superiority of Instantaneous Decisions in Thin Dynamic Matching Markets
University of California Los Angeles(加州大学洛杉矶分校);University of Bristol(布里斯托大学);Université Libre de Bruxelles(布鲁塞尔自由大学);The Institute of Statistical Mathematics(统计数学研究所);The University of Tokyo(东京大学)
Two-Sided Matching Markets in the ELLIS 2020 PhD Program
ELLIS博士项目是一项欧洲计划,通过将优秀的年轻研究人员与AI领域的领先研究人员联系起来,为他们提供支持。特别是,博士生由来自不同国家的两名顾问监督:一名顾问和一名共同顾问。在这项工作中,我们总结了程序,在其最后一步,匹配学生在ELLIS 2020博士课程的顾问。该过程的步骤基于双边匹配市场和大学入学问题的大量文献[Knuth and De Bruijn,1997,Gale and Shapley,1962,Rothand Sotomayor,1992]。我们介绍PolyGS,这是一种适用于双边市场(也称为多对多市场)的算法,我们在预选,面试匹配和最终与顾问匹配的整个选择过程中使用该算法。该算法返回一个稳定的匹配,在这个意义上,没有不匹配的人喜欢被匹配在一起,而不是与他们的当前合作伙伴(给定他们的指示偏好)。Roth [1984]给出的证据表明,随着时间的推移,只有稳定的匹配才有可能被遵守。此外,匹配是学生最佳的。偏好是根据每一方给另一方的排名和研究领域的重叠来构建的。我们提出并讨论了该算法在ELLIS 2020博士课程中产生的匹配。
Dynamic Bipartite Matching Market with Arrivals and Departures
本文研究了二部网络上的一个匹配市场模型,其中每一方的代理人都是按泊松过程随机到达和离开的。对于这样一个动态模型,我们设计了一个机制,不仅决定哪些代理匹配,而且何时匹配他们,以尽量减少不匹配的代理的预期数量。本文的主要贡献是实现不同的定时属性的本地机制的性能的理论界。我们证明了一个等待市场的算法,称为$\textit{Patient}$算法,是指数优于$\textit{Greedy}$算法,即,一种匹配特工的算法这意味着等待在最大化二分网络上的匹配方面有很大的好处。我们注意到,病人的算法需要规划者确定代理人谁是即将离开市场,并在要求下,病人的算法被证明是一个最优的算法。我们还表明,没有要求,贪心算法几乎是最优的。此外,我们考虑了$\textit{1-sided algorithms}$,其中只有一侧的代理可以尝试匹配。这模拟了一个实际的匹配市场,如运费交换市场和劳动力市场,只有一方的代理人可以做出决定。对于这种设置,我们证明了贪婪和耐心算法承认相同的性能,即等待市场是没有价值的。这一结论与双方的代理人都可以做出决定的情况和[Akbarpour等人,$~\textit{Journal of Political Economy}$,2020]。