空间大数据隐私保护机制及模型

Differentially private share and publication of spatial big data have attracted considerable research attention in recent years. The tradeoff among the size of spatial data, data skew, and Laplace noise directly constrains the accuracy of publication. Most existing methods based on tree and grid structure cannot efficiently address the problems because the spatial data is very large scale and skewed. To remedy the deficiency of these solutions, we intend to conduct a framework that employs the sampling, thresholding, sparse vector technology, and filtering on spatial data release under differential privacy in this project. We will propose an efficient kd-tree release method that adopts Bernoulli random sampling and sparse vector technology but completely eliminates the dependency on the depth of the tree. We will design a novel grid-release-based strategy that combines adaptive sampling and high-pass filtering for partitioning spatial data. More specifically, we also propose three-layer adaptive grid for publishing spatial data in terms to different query granularity, and explore the constrained inference techniques to boost query utility. Our results will be used to build a prototype system in order to validate the effectiveness and scalability of our solutions. These studies will provide new ideas, theoretical methods and technical support for further research on differential privacy.

基于差分隐私的空间大数据共享与发布得到了研究者的广泛关注。空间大数据的大小与偏斜性，以及拉普拉斯噪音的多少直接制约着空间数据发布的精度。针对现有基于树结构与网格结构发布方法难以有效兼顾空间数据的大规模性与偏斜性的不足，本项目基于差分隐私保护技术提出集成抽样技术、阈值过滤技术、稀疏向量技术、高通滤波技术的空间数据发布框架。具体拟研究基于伯努利抽样与稀疏向量技术的KD-树发布方法、基于自适应抽样技术与高通滤波技术的网格发布方法、基于阈值过滤的三层自适应网格发布方法，并探讨基于约束推理的查询可用性提升方法。研究成果将用于搭建基于差分隐私的空间大数据共享发布原型系统，以验证所提出发布方法的有效性与高效性。通过本项目的研究为差分隐私保护技术进一步深入研究与应用提供理论方法、技术支撑与新的思路。

空间大数据隐私保护机制及模型已成为隐私保护领域研究的热点。本项目结合空间数据发布与分析带来的系列隐私问题，研究基于中心化/本地化/混洗差分隐私的空间数据发布与分析机制。项目组取得了一些研究成果：发表或录用14篇，投稿1篇，申请授权专利2项。该项目所取得的代表性成果包括：（1）提出了一种基于伯努利随机抽样技术的三层自适应网格空间数据分割方法；（2）提出了一种基于网格分割与四分树索引的空间范围查询响应方法；（3）提出了一种基于序列格且满足中心化差分隐私的空间序列模式挖掘方法；（4）提出了一种满足本地化差分隐私的键-值数据收集与分析方法；（5）提出了一种基于概率联合树的隐私高维数据发布方法。这些研究成果为基于空间大数据隐私保护问题的进一步研究与应用提供了新的解决思路与技术支持。