CAREER: The Algorithmic Foundations of Data Privacy

在过去的十年里，人们越来越依赖数据驱动的技术，包括推荐系统、定向广告和搜索个性化。大数据的这种增长使数据隐私成为一个核心问题。提出的核心问题是：我们如何继续从大型数据集中提取有用的信息，同时可证明地保护这些数据集中包含的个人的一定程度的隐私？这项研究的中心是推进隐私保护数据分析的最新进展。它具体有几个主题：(1)利用被分析的私有数据中的结构，以及分析中使用的查询类，为私有数据分析提供计算高效的算法。(2)深化隐私数据分析与机器学习理论的联系。(3)在数据隐私的基础上放松了大多数工作中隐含的对抗性合谋模型；(4)应用差分隐私工具来有效地利用和分析其他算法环境中的噪声。为了确保这项研究的广泛影响，该项目包括大量的外联活动，包括讲习班组织、课程开发以及编写教科书和其他教育材料。

The past decade has seen a growing reliance on data driven technologies, including recommendation systems, targeted advertising, and search personalization. This growth in big data has made data privacy into a central concern. The central question raised is: how can we continue to extract useful information from large datasets, while provably protecting some measure of privacy for the individuals contained in these datasets?This research centers around advancing the state of the art in privacy preserving data analysis. It specifically has several themes: (1) Exploiting structure in the private data being analyzed, as well as the classes of queries used in the analysis to give computationally efficient algorithms for private data analysis. (2) Deepening the connections between private data analysis and machine learning theory. (3) Relaxing the adversarial collusion model implicit in most work on the foundations of data privacy, and (4) applying the tools of differential privacy to usefully exploit and analyze noise in other algorithmic settings. To ensure the broad impact of this research, this project includes substantial outreach activities, including workshop organization, course development, and the development of a textbook and other educational materials.