The metric distortion of a randomized social choice function (RSCF) quantifies its worst-case approximation ratio to the optimal social cost when the voters' costs for alternatives are given by distances in a metric space. This notion has recently attracted significant attention as numerous RSCFs that aim to minimize the metric distortion have been suggested. Since such tailored voting rules have, however, little normative appeal other than their low metric distortion, we will study the metric distortion of well-established RSCFs. Specifically, we first show that C1 maximal lottery rules, a well-known class of RSCFs, have a metric distortion of $4$, which is optimal within the class of majoritarian RSCFs. Secondly, we conduct extensive computer experiments on the metric distortion of RSCFs to obtain insights into their average-case performance. These computer experiments are based on a new linear program for computing the metric distortion of a lottery and reveal that the average-case metric distortion of some classical RSCFs is often only slightly worse than that of RSCFs tailored to minimize the metric distortion. Finally, we also analytically study the expected metric distortion of RSCFs for the impartial culture distribution. Specifically, we show that, under this distribution, every reasonable RSCF has an expected metric distortion close to $2$ when the number of voters is large.
Smooth Partial Lotteries for Stable Randomized Selection
New Zealand Health Research Council(新西兰健康研究理事会);Swiss National Science Foundation(瑞士国家科学基金会);European Research Council(欧洲研究理事会);Science Foundation Ireland(爱尔兰科学基金会);Volkswagen Foundation(大众基金会);The British Academy(英国学院);Austrian Science Fund (FWF)(奥地利科学基金);Formas(Formas基金会);Luebber et al.(Luebber等人)
本文提出平滑性作为部分彩票设计原则,通过定义评分到选择概率的Lipschitz条件,提出Clipped Linear Lottery机制,证明其在平滑性与遗憾之间取得更好的平衡,并通过实验验证其在实际应用中的有效性。
Sensitivity and Differential Privacy in Metric Voting with Distortion below Three
CyberAgent, Tokyo, Japan(CyberAgent);National Institute of Informatics, Tokyo, Japan(国立情报学研究所);Center for Advanced Intelligence Project, RIKEN, Tokyo, Japan(理化学研究所先进智能研究中心);Kyoto University, Kyoto, Japan(京都大学)
Metric Distortion of Social Choice Rules: Lower Bounds and Fairness Properties
我们在功利扭曲的框架下研究社会选择规则,并对代理人的成本进行了额外的度量假设。在这种方法中,这些成本由所有代理加上替代品的集合上的基础度量给出。社会选择规则只能获得代理人的顺序偏好,而不能获得导致它们的潜在基数成本。失真被定义为社会成本(通常是代理成本的总和)之间的比率选择的替代机制在手,并选择了最佳的替代由一个无所不知的算法。因此,社会选择规则在最坏情况下的扭曲,是衡量在不知道潜在成本的情况下,它总是接近最优选择的程度。在该模型下,已经证实,排名对,众所周知的加权锦标赛规则,达到了最多3的失真[Anshelevich et al. 2015]。我们通过构造一系列实例来证明这一猜想是错误的,这些实例表明排序对的最坏情况失真至少为5。我们对排名对的最坏情况失真的下限与Copeland规则的先前已知的上限相匹配,证明在最坏的情况下,更简单的Copeland规则至少与排名对一样好。只要我们受到(加权或未加权)锦标赛规则的限制,我们就证明随机化无法帮助实现小于3的预期最坏情况失真。在失真框架内使用近似优化的概念,我们证明了Copeland和随机独裁实现了较低的常数因子公平比(分别为5和3),这是对成本和单个最大成本目标的类似结果的相当大的推广。除了上述所有内容之外,我们还概述了该领域进一步研究的几个有趣方向。