SHF: Small: PrivAid: Differentially-Private Analytics for Android Apps

移动应用程序经常使用分析基础设施来收集应用程序性能和用户行为的数据。收集和使用此类数据存在严重的隐私问题。立法努力和社会需求正在呼吁增加透明度，并在数据收集的效用和相应的隐私损失之间达成明确的妥协。该项目将通过使用差异隐私来提高移动应用程序分析的隐私，这为隐私保护分析提供了严格的理论和强大的算法框架。研究和教育的结合将培养移动应用程序和数据分析开发商的专业知识。招收代表性不足的学生将有助于增加计算领域的多样性。该项目将在社会和立法对隐私的需求迅速增长的环境下，推动用户隐私这一重要领域的发展。调查人员将利用局部差异隐私(LDP)理论，在移动应用分析中开发严格和可量化的隐私机制。该项目的第一个贡献将是Privaid的设计和实施，这是一种概念性方法和相关软件工具，用于在广泛使用的移动应用分析基础设施中部署LDP分析。在此基础上，将利用新的随机化和抽样技术开发几种针对重要软件分析的算法：(1)控制流图中的边/路径分析；(2)用于事件分析和图分析的重量级分析；(3)连续观察下的分析。Privaid、软件分析和实验对象的公开发布将使未来LDP软件分析的开发和评估成为可能。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Mobile apps often use analytics infrastructures to gather data on appperformance and user behaviors. There are significant privacy concernsabout the collection and use of such data. Legislative efforts andsocietal demands are calling for increased transparency andwell-defined compromises between the utility of data gathering and thecorresponding loss of privacy. This project will improve the privacyof mobile app analytics by employing differential privacy, whichprovides a rigorous theory and a powerful algorithmic framework forprivacy-preserving analysis. Integration of researchand education will develop the expertise of developers of mobile appsand data analyses. Recruitment of underrepresented students willcontribute to increased diversity in computing. The project willadvance the important area of user privacy, in an environment withrapidly-increasing societal and legislative demands for privacy.The investigators will develop rigorous and quantifiable privacymechanisms in mobile-app analytics by leveraging the theory of localdifferential privacy (LDP). The first contribution of the project willbe the design and implementation of PrivAid, a conceptual approach andrelated software tools for deploying LDP analyses in widely-usedinfrastructures for mobile-app analytics. Based on this foundation,several algorithms targeting important software analyses will bedeveloped using novel techniques for randomization and sampling: (1)edge/path profiling in control-flow graphs; (2) analyses of heavyhitters for event profiling and graph profiling; (3) profiling undercontinuous observation. The public release of PrivAid, softwareanalyses, and experimental subjects will enable the development andevaluation of future LDP software analyses.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.