SaTC: CORE: Small: New Techniques for Optimizing Accuracy in Differential Privacy Applications

SaTC：核心：小型：优化差异隐私应用准确性的新技术

• 批准号: 1931686
• 负责人: Daniel Kifer
• 金额: $ 50万
• 依托单位: Pennsylvania State Univ University Park
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1931686&HistoricalAwards=false
• 关键词: SaTC; CORE; Small; New; Techniques

Differential Privacy is an important advance in the modern toolkit for protecting privacy and confidentiality. It allows organizations such as government agencies and private companies to collect data and publish statistics about it without leaking personal information about people -- no matter how sophisticated an attacker is. The project's novelties are in the careful design of new differentially private tools that provide more accurate population statistics while maintaining strong privacy guarantees. The project's impacts are in the ability to create datasets for social science and policy research without sacrificing privacy of individuals. The project includes both graduate and undergraduate students in this research. The technical ideas behind this project are that a careful analysis of privacy proofs for many differentially private algorithms can identify additional (noisy) information that can be released without changing the privacy guarantees. In addition to this, noise that is correlated between different stages of a differentially private algorithm can further reduce the variance of the final result. These techniques will allow smaller organizations to optimize the accuracy of their privacy preserving algorithms for practical deployment, and customize existing algorithms by swapping in building blocks that have more accuracy and that allow them to take better advantage of background knowledge.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.

差分隐私是保护隐私和机密性的现代工具包中的一个重要进步。它允许政府机构和私营公司等组织收集数据并发布有关数据的统计数据，而不会泄露有关人员的个人信息-无论攻击者多么复杂。该项目的新颖之处在于精心设计了新的差异化隐私工具，这些工具提供了更准确的人口统计数据，同时保持了强有力的隐私保障。该项目的影响在于能够在不牺牲个人隐私的情况下为社会科学和政策研究创建数据集。该项目包括研究生和本科生在这项研究。 这个项目背后的技术思想是，仔细分析许多差异隐私算法的隐私证明可以识别额外的（噪音）信息，这些信息可以在不改变隐私保证的情况下发布。除此之外，在差分隐私算法的不同阶段之间相关的噪声可以进一步减小最终结果的方差。这些技术将允许较小的组织优化其隐私保护算法的准确性，以进行实际部署，并通过交换具有更高准确性的构建块来定制现有算法，使其能够更好地利用背景知识。该奖项反映了NSF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进行评估，被认为值得支持的搜索.

项目成果

Scaling up Differentially Private Deep Learning with Fast Per-Example Gradient Clipping

• DOI: 10.2478/popets-2021-0008
• 发表时间: 2020-09
• 期刊: Proceedings on Privacy Enhancing Technologies
• 作者: Jaewoo Lee;Daniel Kifer

Free gap estimates from the exponential mechanism, sparse vector, noisy max and related algorithms

来自指数机制、稀疏向量、噪声最大值和相关算法的自由间隙估计

• DOI: 10.1007/s00778-022-00728-2
• 发表时间: 2022
• 期刊: The VLDB Journal
• 作者: Ding, Zeyu;Wang, Yuxin;Xiao, Yingtai;Wang, Guanhong;Zhang, Danfeng;Kifer, Daniel
• 通讯作者: Kifer, Daniel

Answering Private Linear Queries Adaptively Using the Common Mechanism

使用通用机制自适应地回答私有线性查询

• DOI: 10.14778/3594512.3594519
• 发表时间: 2023
• 期刊: Proceedings of the VLDB Endowment
• 影响因子: 2.5
• 作者: Xiao, Yingtai;Wang, Guanhong;Zhang, Danfeng;Kifer, Daniel
• 通讯作者: Kifer, Daniel

Optimizing Fitness-For-Use of Differentially Private Linear Queries

• DOI: 10.14778/3467861.3467864
• 发表时间: 2020-11
• 期刊: ArXiv
• 作者: Yingtai Xiao;Zeyu Ding;Yuxin Wang;Danfeng Zhang;Daniel Kifer

DPGen: Automated Program Synthesis for Differential Privacy

DPGen：差异隐私的自动程序合成

• DOI: 10.1145/3460120.3484781
• 发表时间: 2021
• 期刊: Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security
• 作者: Wang, Yuxin;Ding, Zeyu;Xiao, Yingtai;Kifer, Daniel;Zhang, Danfeng
• 通讯作者: Zhang, Danfeng