NeTS: Small: Protecting Privacy While Providing Utility in Published Network Mobility Traces Using Differential Privacy

记录的个人在 WiFi 网络中的接入点之间或蜂窝网络中的蜂窝塔之间移动的信息通常被视为私人数据，不会被网络运营商泄露。 然而，现实世界中个人移动的这种痕迹对于研究人员和工程师评估新的移动网络协议、架构和应用程序至关重要。 该项目将开发和评估用于操纵和发布移动轨迹的技术，以便（i）向其数据包含在原始轨迹中的用户提供明确定义的、经过正式证明的隐私保证，以及（ii）发布的轨迹在用于常见计算机网络和协议设计和分析任务时提供高精度。 该项目中开发的差分隐私技术使用原始数据的约束轨迹前缀表示来确定原始数据的底层表示，并明智地分配满足差分隐私（DP）约束所需的随机噪声。 导出的 DP 轨迹的效用（使用导出的 DP 轨迹实现的特定任务性能与使用原始轨迹实现的任务特定性能之间的差异）将针对四个常见且有代表性的用例进行评估。原始数据本身将从运行的 4,500 个节点 WiFi 校园无线网络中获取。由于更受限制的输出 DP 数据可能会为给定隐私预算的特定任务提供更高的效用（比 DP 跟踪），因此该项目还将检查针对特定任务的专门 DP 输出的效用权衡。 该项目还将开发已发布的 DP 数据的误差范围，以便为网络分析师提供与 DP 数据相关的“随机性”程度的定量测量。该项目允许网络研究人员和分析师获得“现实世界”的网络移动数据，同时向数据中代表其活动的用户提供正式的隐私保证。 作为该项目的一部分公开提供的用户移动性的具体 DP 痕迹也将引起人们的兴趣和使用，随后围绕最合适和最有用的私人移动性数据形式的社区讨论也将引起人们的兴趣和使用。更广泛地说，在“大数据”的新生时代，隐私已变得至关重要。移动无线用户及其在网络接入点之间的轨迹只是轨迹足迹的一个示例 - 访问的物理位置序列、遇到的熟人、访问的网站或任何无数其他活动都是远远超出网络研究界的重大兴趣的“轨迹”。因此，本项目中进行的特定领域研究可能会在这些领域和许多其他领域得到应用。对研究生研究助理和本科生 REU 学生的指导，以及将隐私相关的研究成果纳入研究生水平的研讨会，以及（更广泛地）将隐私问题纳入本科生网络课程，将为该项目的研究带来更广泛的影响。

Recorded information of an individual's movement among access points in a WiFi network or among cell towers in a cellular network is typically regarded as private data and not divulged by network operators. Yet such traces of real-world individual mobility are critically important for researchers and engineers in evaluating new mobile network protocols, architectures, and applications. This project will develop and evaluate techniques for manipulating and then publishing mobility traces in a manner such that (i) well-defined, formally-proven, privacy guarantees are provided to the users whose data are contained in the original traces, and (ii) the published traces provide high accuracy when used in common computer network and protocol design and analysis tasks. The differential privacy techniques developed in this project use a constrained trajectory-prefix representation of the original data to determine the underlying representation of the original data and judiciously allocate random noise needed to satisfy differential privacy (DP) constraints. The utility of the derived DP trajectories - the difference between task-specific performance realized using the derived DP trajectories and realized using the original trajectories - will be evaluated for four common and representative use cases. The original data itself will be obtained from an operational 4,500-node WiFi campus wireless network. Since a more restricted output DP data may provide higher utility (than DP traces) for a specific task for a given privacy budget, this project will also examine utility tradeoffs from specializing DP output for specific tasks. The project will also develop error bounds on published DP data to provide the network analyst with a quantitative measure of the degree of "randomness" associated with the DP data. This project allows 'real world' network mobility data to be made available to networking researchers and analysts, while at the same time providing formal privacy guarantees to users whose activities are represented in that data. The specific DP traces of user mobility made publicly available as part of this project will also be of interest and use, as will the ensuing community discussion around the most appropriate and useful forms of private mobility data. More broadly, in the nascent era of "big data," privacy has become of paramount importance. Mobile wireless users and their trajectories among network access points are just one example of trajectory footprints - sequences of physical locations visited, acquaintances met, web sites visited or any of a myriad of other activities are 'trajectories' of significant interest well beyond the networking research community. The domain-specific research undertaken in this project may thus find application in these and many other areas. The mentoring of a diverse group of graduate research assistants and undergraduate REU students and incorporation of privacy-related research results into graduate-level seminars and (more broadly) privacy issues into undergraduate network curriculum will provide additional broader impact for this project's research.