Collaborative Research: SaTC: CORE: Small: Privacy and Fairness in Critical Decision Making

许多机构或公司发布关于个人群体的统计数据，然后将其用作关键决策过程的输入。例如，人口普查数据用于向各州和辖区分配资金和关键资源。由此产生的决策可以对参与的个人产生重大的社会和经济影响。在许多情况下，发布的数据包含敏感信息，这些信息的隐私受到严格的监管，差分隐私（DP）已成为数据隐私保护的首选范式。然而，虽然差别隐私为公布的数据提供了强有力的隐私保障，但最近很明显，它可能在下游决策过程中引起偏见和公平问题，包括联邦资金的分配、国会席位的分配以及疫苗和治疗方法的分配。这些偏见和公平问题可能会对许多人的健康、福祉和归属感产生不利影响，而且人们对这些问题知之甚少。这个项目在隐私、公平、偏见和决策过程的交叉点解决了这一知识差距。它将提供不同隐私工具的新视角，以共同解决关键决策过程中的公平和隐私问题。它将量化这些应用中产生的不同影响，并提供新的机制和缓解技术，以克服其中一些问题。这些贡献将嵌入到建模和软件工具中，以使该技术广泛可用和适用。从科学的角度来看，该项目将开发新一代的隐私保护工具，通过利用来自不同隐私、优化和编程语言的知识，将解决其设计中的偏见和公平问题，而不是事后的想法。该项目在五个方向上贡献了新的科学知识：(1)它识别并理解了当使用DP数据发布时可能受到公平性问题影响的下游决策过程的结构；(2)识别并理解可能引入偏差的DP机制的结构；(3)定义了对偏见和公平问题进行表征和推理的理论框架；(4)设计缓解措施，消除或减轻偏见和公平问题，在隐私、准确性和公平性之间找到适当的权衡；(5)开发建模和软件工具，以自动识别和解释偏见和公平问题，并从决策过程的规范中派生缓解措施。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Many agencies or companies release statistics about groups of individuals that are then used as input to critical decision processes. For example, census data is used to allocate funds and distribute critical resources to states and jurisdictions. The resulting decisions can have significant societal and economic impacts for participating individuals. In many cases, the released data contain sensitive information whose privacy is strictly regulated and Differential Privacy (DP) has become the paradigm of choice for protecting data privacy. However, while differential privacy provides strong privacy guarantees on the released data, it has become apparent recently that it may induce biases and fairness issues in downstream decision processes, including the allotment of federal funds, apportionment of congressional seats, and distribution of vaccines and therapeutics. These biases and fairness issues may adversely affect the health, well-being, and sense of belonging of many individuals, and are poorly understood. This project addresses this knowledge gap at the intersection of privacy, fairness, bias, and decision processes. It will offer novel perspectives on differential privacy tools to address fairness and privacy jointly in critical decision processes. It will quantify the disparate impact arising in these applications and contribute novel mechanisms and mitigation techniques to overcome some of these issues. These contributions will be embedded in modeling and software tools to make the technology widely available and applicable.From a scientific standpoint, this project will develop a new generation of privacy-preserving tools that, by exploiting knowledge from differential privacy, optimization, and programming languages, will address biases and fairness issues in their designs, not as an afterthought. The project contributes new scientific knowledge along with five directions: (1) it identifies and understands the structure of downstream decision processes that may be subject to fairness issues when using DP data releases; (2) it identifies and understands the structure of DP mechanisms that may introduce biases; (3) it defines theoretical frameworks to characterize and reason about biases and fairness issues; (4) it designs mitigation measures that would remove or alleviate the biases and fairness issues, finding an appropriate tradeoff between privacy, accuracy, and fairness; (5) it develops modeling and software tools to automatically identify and explain biases and fairness issues, and derive mitigation measures from the specification of the decision process.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.