面向众包计算的差分隐私保护机制与算法研究

Although crowdsourcing has many applications, the users’ privacy leakage in crowdsourcing has become one of the major problems that hamper the further popularization of crowdsourcing applications. In contrast to the privacy-reserving problems appeared in traditional database systems, the study on the privacy-preserving problems in crowdsourcing should take into account a series of unique features of the crowdsourcing paradigm, including the users’ laziness, strategic and dynamic behaviors, and heterogeneity. Facing these unique features, this project studies various differential privacy-preserving problems concerning the users’ personal attribute information, status information and data information, by leveraging the concept and theory of differential privacy. The detailed issues studied in this project include the differentially private pricing mechanism for dynamic users, the differentially private and budget-feasible auction mechanisms, the joint differentially private algorithms for spatial crowdsourcing task assignment, the spatial crowdsourcing algorithms with privacy protection on the users’ moving trajectories, the statistical crowdsourcing algorithms based on the local differential privacy model and the differentially private statistical algorithms with dynamic data acquisitions. Through the study of this project, we will design a series of practical and efficient differentially private mechanisms and algorithms for crowdsourcing, which can provide significant theoretical and technical supports to improve the practicability and popularity of the crowdsourcing paradigm.

众包计算具有广泛的应用，但众包用户的隐私泄露问题是阻碍众包计算应用进一步普及的关键问题之一。与传统数据库系统的隐私保护问题不同，众包计算中的隐私保护研究需考虑一系列众包计算的独有特征，包括众包用户的怠惰性、策略性、动态性和异构性等。为此，本项目基于差分隐私概念和理论，并针对众包计算的独有特征，对涉及众包用户个人属性、状态属性、数据属性等各类信息的差分隐私保护问题进行研究；具体研究内容包括面向动态用户群体的差分隐私定价机制、具有预算约束的差分隐私拍卖机制、基于共同差分隐私模型的地理众包任务分配算法、保护用户运动轨迹隐私的地理众包算法、基于本地差分隐私模型的众包数据统计算法和动态数据获取环境下保护差分隐私的统计算法等。通过这一研究，本项目将设计出一系列高效实用的众包计算的差分隐私保护机制和算法，能够为促进众包计算的普及性和实用性提供有效的理论和技术支撑。

众包计算具有广泛的应用，但众包用户的隐私泄露问题是阻碍众包计算应用进一步普及的关键问题之一。本项目基于差分隐私概念和理论，并针对众包计算中用户的怠惰性、策略性、动态性和异构性等独有特征，设计出一系列保护众包用户隐私的机制与算法。在保护众包用户代价隐私的激励机制设计方面，基于拍卖理论、多臂老虎机理论等提出了离线和在线两种情况下的符合预算约束的定价机制，除证明了所提机制符合差分隐私保护的要求之外，还证明了其近似比、悔过、诚实性、个体理性等一系列性能指标；在保护众包用户地理位置和轨迹隐私方面，提出了新型的地理众包任务分配框架，并通过众包用户坐标变换、加入拉普拉斯噪声等手段有效保护用户隐私；在保护众包用户数据隐私的统计算法方面，针对高维众包数据集提出了保护差分隐私的频繁项集挖掘算法，并针对大规模众包数据集提出了低复杂度的保护差分隐私的次模数据约简算法。大量实验结果表明，项目所提出的机制与算法在精度和效率等多项指标上显著优于已有算法。本项目的研究能够为促进众包计算的安全性及普及性提供有效的理论和技术支撑，相关研究成果已经发表在INFORMS Journal on Computing、IEEE Transactions on Mobile Computing、IEEE Trans. on Networking、ICML、SIGMETRICS等多种国际顶级期刊和会议上。在项目支持下，项目组目前已培养博士、硕士研究生20余人，并多次获得研究生国家奖学金。