Full Citation

Title: Differentially Private Learning with Small Public Data

Citation Type: Conference Paper

Publication Year: 2020

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: Differentially private learning tackles tasks where the data are private and the learning process is subject to differential privacy requirements. In real applications, however, some public data are generally available in addition to private data, and it is interesting to consider how to exploit them. In this paper, we study a common situation where a small amount of public data can be used when solving the Empirical Risk Minimization problem over a private database. Specifically, we propose Private-Public Stochastic Gradient Descent, which utilizes such public information to adjust parameters in differentially private stochastic gradient descent and fine-tunes the final result with model reuse. Our method keeps differential privacy for the private database, and empirical study validates its superiority compared with existing approaches.

Url: https://cs.nju.edu.cn/zhouzh/zhouzh.files/publication/aaai20ppsgd.pdf

Url: https://cs.nju.edu.cn/zhouzh/zhouzh.files/publication/publication.htm

User Submitted?: No

Authors: Wang, Jun; Zhou, Zhi-Hua

Conference Name: Conference on Artificial Intelligence

Publisher Location: New York, New York

Data Collections: IPUMS USA

Topics: Other, Population Data Science

Countries: United States