EAGER: Toward Transparency in Public Policy via Privacy-Enhanced Social Flow Analysis with Applications to Ecological Networks and Crime

最近在计算能力、数据收集和数据科学方面的改进使科学数据分析取得了巨大进步。然而，相关数据往往是高度敏感的(例如，人口普查记录、税务记录、医疗记录)。该项目解决了一个新出现的关键科学问题：隐私问题限制了对可能泄露个人信息的原始数据的访问。对这些数据进行“消毒”的技术(例如，匿名化)可能会对使用这些数据的科学结果的质量产生负面影响。我们如何提供既能保护个人隐私又能准确支持科学分析的数据？该项目解决了保护隐私的净化数据分析方面的挑战：(1)如何分析净化数据，以使结论经得起同行审查？(2)哪些工作流程和可视化必须得到隐私技术的支持？(3)在没有原始数据的情况下，科学家如何评估净化带来的偏见？该项目特别侧重于“社会流动分析”，即对个人或家庭的敏感社会流动数据(例如，通勤模式、迁移轨迹)进行数据分析。研究人员正在开发一个社区网络的生态模型，这些社区网络通过社会流动联系在一起，并研究社会流动是如何形成和维持的。该项目正在对开发此类理论所需的数据访问和可视化类型进行分类，研究既可扩展又具有统计稳健性的替代分析，开发初步的隐私保护数据保护方法，并评估隐私保护方法是否能够得出与获取原始数据相同的结论。

Recent improvements in computing capabilities, data collection, and data science have enabled tremendous advances in scientific data analysis. However, the relevant data are often highly sensitive (e.g., Census records, tax records, medical records). This project addresses an emerging and critical scientific problem: Privacy concerns limit access to raw data that might reveal information about individuals. Techniques to "sanitize" such data (e.g., anonymization) could have negative impact on the quality of the scientific results that use the data. How can we provide data that protect the privacy of individuals but also accurately support scientific analyses? The project addresses challenges regarding analysis of privacy-preserving sanitized data: (1) How can sanitized data be analyzed so that conclusions will stand up to peer review? (2) What workflows and visualizations must be supported by privacy technology? (3) How can scientists assess bias introduced by sanitization without access to the raw data? The project focuses specifically on "social flow analysis," in which data analysis is performed on sensitive social flow data (e.g., commuting patterns, migration trajectories) of individuals or families. The researchers are developing an ecological model of networks of neighborhoods that are linked by social flows and studying how social flows are formed and maintained. The project is cataloging the types of data access and visualization needed to develop such theories, studying alternative analyses that are both scalable and statistically robust, developing preliminary privacy-preserving data protection methods, and evaluating whether the privacy-preserving methods enable the same conclusions as access to the raw data.