Differentially private knowledge transfer for federated learning

Tao Qi, Fangzhao Wu (), Chuhan Wu (), Liang He, Yongfeng Huang () and Xing Xie
Additional contact information
Tao Qi: Tsinghua University
Fangzhao Wu: Microsoft Research Asia
Chuhan Wu: Tsinghua University
Liang He: Tsinghua University
Yongfeng Huang: Tsinghua University
Xing Xie: Microsoft Research Asia

Nature Communications, 2023, vol. 14, issue 1, 1-9

Abstract: Abstract Extracting useful knowledge from big data is important for machine learning. When data is privacy-sensitive and cannot be directly collected, federated learning is a promising option that extracts knowledge from decentralized data by learning and exchanging model parameters, rather than raw data. However, model parameters may encode not only non-private knowledge but also private information of local data, thereby transferring knowledge via model parameters is not privacy-secure. Here, we present a knowledge transfer method named PrivateKT, which uses actively selected small public data to transfer high-quality knowledge in federated learning with privacy guarantees. We verify PrivateKT on three different datasets, and results show that PrivateKT can maximally reduce 84% of the performance gap between centralized learning and existing federated learning methods under strict differential privacy restrictions. PrivateKT provides a potential direction to effective and privacy-preserving knowledge transfer in machine intelligent systems.

Date: 2023
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (1)

Downloads: (external link)
https://www.nature.com/articles/s41467-023-38794-x Abstract (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:nat:natcom:v:14:y:2023:i:1:d:10.1038_s41467-023-38794-x

Ordering information: This journal article can be ordered from
https://www.nature.com/ncomms/

DOI: 10.1038/s41467-023-38794-x

Access Statistics for this article

Nature Communications is currently edited by Nathalie Le Bot, Enda Bergin and Fiona Gillespie

More articles in Nature Communications from Nature
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().