泛研网

公众版 查看说明
全球科研项目数据库2026版

CAREER:Exact Optimal and Data-Adaptive Algorithms and Tools for Differential Privacy

项目来源

美国国家科学基金(NSF)

项目主持人

Yu-Xiang Wang

项目受资助机构

UNIVERSITY OF CALIFORNIA SAN DIEGO

财政年度

2025,2020

立项时间

未公开

项目编号

2503856

项目级别

国家级

研究期限

未知 / 未知

受资助金额

509630.00美元

学科

未公开

学科代码

未公开

基金类别

Continuing grant

关键词

Secure&Trustworthy Cyberspace ; SaTC:Secure and Trustworthy Cyberspace ; CAREER-Faculty Erly Career Dev

参与者

未公开

参与机构

UNIVERSITY OF CALIFORNIA,SAN DIEGO

项目标书摘要：This project is motivated by the increasing public concerns on privacy issues,new legislations and the high demand for privacy enhancing technologies such as differential privacy(DP)in applications from both private and public sectors.The overarching theme of the project is to address the pressing new challenges that arise as differential privacy transforms from a theoretical construct into a practical technology.The project advances the state-of-the-art of research in the area of DP,and contributes to privacy education.On the research front,the project develops new algorithms and analytical tools that enable more precise privacy accounting and higher utility in DP.On the education front,the project involves training future leaders in DP areas,creating educational materials and expanding an open-source software library called autodp that makes state-of-the-art differentially private computation more accessible.Collectively,the integrated research and educational activities contribute to ongoing collaborative efforts in building innovative applications of differential privacy.The project has three main components in use-inspired fundamental research.The first component unifies the recent breakthroughs in DP,such as,Renyi DP,moments accountant,f-DP and produce an intermediate functional representation that allows lossless conversions among these representations.The second component focuses on investigating the stronger privacy properties permitted by the structures of the actual data,and addressing the dilemma of interpreting worst-case privacy on average-case data.The third component focuses on using a public dataset to``denoise''the private data releases or to facilitate private machine learning.The outputs of the research will be broadly shared through integration in autodp library,and will be integrated in courses.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.

人员信息

Yu-Xiang Wang(Principal Investigator)：yuxiangw@ucsd.edu；

机构信息

【University of California-San Diego(Performance Institution)】StreetAddress：9500 GILMAN DR,LA JOLLA,California,United States/ZipCode：920930934；【UNIVERSITY OF CALIFORNIA,SAN DIEGO】StreetAddress：9500 GILMAN DR,LA JOLLA,California,United States/PhoneNumber：8585344896/ZipCode：920930021；

项目主管部门

Directorate for Computer and Information Science and Engineering(CSE)-Division Of Computer and Network Systems(CNS)

项目官员

Xiaogang(Cliff)Wang(Email：xiawang@nsf.gov；Phone：7032922812)

  • 排序方式：
  • 1
  • /

  • 1.Revisiting model-agnostic private learning: Faster rates and active learning

    • 关键词：
    • Artificial intelligence ; Learning systems;Active Learning ; Differential privacies ; Faster rates ; Learn+ ; Model;agnostic private learning ; Noise conditions ; Private aggregation of teacher ensemble ; Private aggregations ; Teachers' ; Tsybakov noise condition
    • LiuChong;ZhuYuqing;ChaudhuriKamalika;WangYu-Xiang
    • 《Journal of Machine Learning Research》
    • 2021年
    • 22卷
    • 期刊

    The Private Aggregation of Teacher Ensembles (PATE) framework is one of the most promising recent approaches in differentially private learning. Existing theoretical analysis shows that PATE consistently learns any VC-classes in the realizable setting, but falls short in explaining its success in more general cases where the error rate of the optimal classifier is bounded away from zero. We fill in this gap by introducing the Tsybakov Noise Condition (TNC) and establish stronger and more interpretable learning bounds. These bounds provide new insights into when PATE works and improve over existing results even in the narrower realizable setting. We also investigate the compelling idea of using active learning for saving privacy budget, and empirical studies show the effectiveness of this new idea. The novel components in the proofs include a more refined analysis of the majority voting classifier — which could be of independent interest — and an observation that the synthetic "student" learning problem is nearly realizable by construction under the Tsybakov noise condition. ©2021 Chong Liu, Yuqing Zhu, Kamalika Chaudhuri, and Yu-Xiang Wang.

    ...
  • 排序方式：
  • 1
  • /