Collaborative Research: SaTC: CORE: Medium: Differentially Private SQL with flexible privacy modeling, machine-checked system design, and accuracy optimization

公司和政府机构维护着对其运营至关重要的大型数据库。这类数据库包含有关人们与州和地方机构的互动(例如，税务申报、旅行数据)或与公司的互动(例如，客户档案和购买历史、员工工资和税务数据以及绩效评估)的敏感信息。然而，这样的数据库也具有巨大的分析价值，可以用来改善内部运作，指导政策决策，并提供关于社会的综合信息。“正式隐私”是一个科学领域，研究如何在分析中注入噪音以保护机密信息，而不会对分析的效用造成不利影响。然而，现有技术很难应用，需要大量的技术专长。这个项目的目标以及更广泛的意义和重要性是使人们能够民主化地使用先进的正式隐私工具。该项目的创新之处在于：(1)一个可定制的隐私模型，用于捕获数据库中不同的隐私问题；(2)自动化工具，用于推断必须向数据分析注入多少噪声才能满足这些隐私问题，而不会对分析结果产生不利影响。以前的工作使用简单的、预先指定的隐私模型，这严重限制了可以支持的应用程序类型，并且在设计这些系统时需要大量的技术专业知识才能获得准确的查询答案。项目组开发了一个用于SQL数据库的中间件应用程序，包括(1)用于分析数据库模式和交互开发隐私模型的自动化工具，其中数据元素需要差异隐私变化的似是而非的否认；以及(2)用于对SQL查询进行推理并将隐私保护查询执行计划定制为最适合数据的隐私模型的自动化工具。最终的结果是一个开源的、可定制的、隐私保护的数据库分析系统，与现有的SQL数据库兼容。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。

Companies and government agencies maintain large databases crucial to their operations. Such databases contain sensitive information about people's interactions with state and local agencies (e.g., tax filings, travel data) or interactions with companies (e.g., customer profiles and purchase histories, employee salary and tax data, and performance reviews). However, such databases also have immense value for analytics that can be used to improve internal operations, guide policy decisions, and provide aggregate information about society. "Formal Privacy" is a scientific field that studies how to inject noise into analyses to protect confidential information without adversely affecting the utility of the analyses. However, existing technology is difficult to apply and requires significant technical expertise. The goal, and broader significance and importance of this project are to democratize access to advanced formal privacy tools. The project's novelties are (1) a customizable privacy model for capturing different privacy concerns in a database and (2) automated tools that reason about how much noise must be injected into a data analysis to satisfy these confidentiality concerns without adversely affecting the analysis results. Prior work used simple, pre-specified privacy models that severely limited the types of applications that can be supported and required significant technical expertise in the design of those systems to obtain accurate query answers. The project team develops a middleware application for SQL databases consisting of (1) automated tools for analyzing a database schema and interactively developing a privacy model of which data elements need the plausible deniability of differential privacy variations and (2) automated tools for reasoning about SQL queries and customize privacy-preserving query execution plans to the privacy model that is most appropriate for the data. The end result is an open-source, customizable, privacy-preserving database analytics system compatible with existing SQL databases.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.