【作者】 周宇；

【导师】 仲盛； 张渊；

【作者基本信息】 南京大学 ， 计算机科学与技术， 2018， 硕士

【摘要】 现如今,随着智能便携设备(如智能手机,可穿戴设备等)的普及,移动群体感知(Mobile Crowd Sensing,MCS)逐渐成为一种收集大规模传感器数据的有效模式。与传统的无线传感器网络相比,MCS可以节约大量的人力和物力。另外,MCS也逐渐开始应用到个人的日常生活中,例如医疗健康,智慧交通等。现如今MCS还有一些重要的问题亟待人们去解决,包括:保证数据来源、保证数据的可靠性、保护数据的隐私安全。现有的研究工作在解决以上问题的同时,并没有考虑到平台预算和用户提供的数据同时有限的情况。因此,如何设计一个综合多方面考虑的MCS系统的框架,使得其在以上多种限制条件下,收集到尽可能多可靠的数据具有重要的研究意义。本文设计了基于反向拍卖的激励机制,它考虑到了平台预算和用户提供的数据的有限性,并可以激励用户参与MCS系统,同时还可以保证用户诚实地提供自已的数据,从而使得平台可以获得尽可能多的可靠的数据。另外,本文还使用了差分隐私的技术,保护了用户的数据隐私。注意到平台通常被认为是可靠的,本文将分两种情况对用户进行讨论,一种是所有用户都对别的用户的数据不感兴趣,另一种是部分用户是潜在的攻击者,他们对别的用户的数据很感兴趣。前者没有隐私泄露的隐患,本文设计一个真实的、个人理性的、预算可行的和效益最大化的拍卖机制;而对于后者,本文设计了一个近似真实的、个人理性的、预算可行的和效益最大化的拍卖机制,同时,该拍卖机制满足差分隐私协议,可以保护用户的隐私不被别的用户所得到。本文通过大量的实验验证了拍卖机制的性能,结果表明设计的拍卖机制可以使平台获得的数据量近似最大化,并能有效保证用户的隐私安全。更多还原

【Abstract】 With the popularization of human-carried devices(such as smart phone,wearable device),it is possible for Mobile Crowd Sensing(MCS)applications to perform large scale sensing with sensors embedded in these devices.Due to the much fewer sens-ing costs and much higher sensing coverage compared to traditional sensing networks,there are many application scenarios in our daily life concerning healthcare,smart trans-portation and so on.Nowadays,there are still some important issues in the MCS that need to be resolved,including ensuring data sources,ensuring data reliability and pro-tecting data privacy.However,many researchers did not consider the limited nature of platform budgets and user-provided data.Therefore,designing an MCS system with comprehensive considerations which can collect as much reliable data as possible has important research significance.In the paper,we design incentive mechanisms based on reverse auction which ensure approximately maximized value of services provided by selected workers and ensure honest bids submitted by workers.Meanwhile,these two mechanisms consider situations where there are constraints on both crowdsourcer’s budget and workers’ ca-pacity.In addition,we also use differential privacy technology to protect users’ data privacy.Not only do we study the scenario where every worker is trusted and will not infer others’ bids,but we also investigate the scenario where there are some honest-but-curious workers.For the former,we design a truthful,individual rational,and computationally efficient incentive mechanism that achieves nearly optimal benefits.For the latter,we design an approximately truthful,individual rational,computationally efficient,and differentially private incentive mechanism that helps to protect workers’privacy from the infringement of curious workers and achieves nearly optimal benefits.Rigorous theoretical analyses and extensive simulations are given to validate the above properties and evaluate the performance of our incentive mechanisms.更多还原