Personalized differential privacy approach in the federated learning scenario

Zhenzhou Guo; Rongchang Dong; Ao Liu

doi:10.1117/12.3024652

3 April 2024 Personalized differential privacy approach in the federated learning scenario

Zhenzhou Guo, Rongchang Dong, Ao Liu

Proceedings Volume 13078, Second International Conference on Informatics, Networking, and Computing (ICINC 2023); 130780F (2024) https://doi.org/10.1117/12.3024652
Event: Second International Conference on Informatics, Networking, and Computing (ICINC 2023), 2023, Wuhan, China

Abstract

In order to deal with the problem of model gradients, neural network weights and other parameters exposing user privacy information and user privacy personalization under federated learning, therefore, this article proposes a personalized differential privacy method in a federated learning scenario. Before training, users upload the privacy protection level to the central server. The central server defines the amount of noise based on the total privacy budget and privacy protection level, and aggregates the training parameters using an aggregation algorithm weighted by the privacy protection level; the algorithm is verified to be effective in MNIST and CIFAR-10 data set and compared with existing localized differential privacy algorithms and centralized differential privacy algorithms, The classification accuracy is 94.12% and 45.20% respectively. Proved the usability and rationality of the algorithm in this scenario.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Zhenzhou Guo, Rongchang Dong, and Ao Liu "Personalized differential privacy approach in the federated learning scenario", Proc. SPIE 13078, Second International Conference on Informatics, Networking, and Computing (ICINC 2023), 130780F (3 April 2024); https://doi.org/10.1117/12.3024652

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available