On the Privacy Leakage of Over-the-Air Federated Learning Over MIMO Fading Channels DOI
Hang Liu, Yan Jia, Ying–Jun Angela Zhang

et al.

GLOBECOM 2022 - 2022 IEEE Global Communications Conference, Journal Year: 2023, Volume and Issue: unknown, P. 5274 - 5279

Published: Dec. 4, 2023

Federated learning (FL) allows edge devices to collaboratively train machine models without directly sharing data. While over-the-air model aggregation improves communication efficiency, uploading can lead privacy risks. Previous research has focused on FL with a single-antenna server, leveraging noise enhance user-level privacy. This method achieves the so-called "free" by decreasing transmit power instead of introducing additional privacy-preserving mechanisms at devices. In this paper, we analyze leakage over multiple-input multiple-output (MIMO) fading channel. We show that multiple-antenna server amplifies leakage. Consequently, relying solely is inefficient meet high requirements, particularly when receive antenna array large. calls for joint optimization algorithm device-side mechanism and receiving protocol achieve better privacy-learning tradeoff. Numerical results validate our analysis highlight impact size

Language: Английский

Convergence-Privacy-Fairness Trade-Off in Personalized Federated Learning DOI Creative Commons
Xiaofang Zhao, Qimei Cui, Weicai Li

et al.

IEEE Transactions on Machine Learning in Communications and Networking, Journal Year: 2025, Volume and Issue: 3, P. 246 - 262

Published: Jan. 1, 2025

Language: Английский

Citations

1
On the Privacy Leakage of Over-the-Air Federated Learning Over MIMO Fading Channels DOI
Hang Liu, Yan Jia, Ying–Jun Angela Zhang

et al.

GLOBECOM 2022 - 2022 IEEE Global Communications Conference, Journal Year: 2023, Volume and Issue: unknown, P. 5274 - 5279

Published: Dec. 4, 2023

Federated learning (FL) allows edge devices to collaboratively train machine models without directly sharing data. While over-the-air model aggregation improves communication efficiency, uploading can lead privacy risks. Previous research has focused on FL with a single-antenna server, leveraging noise enhance user-level privacy. This method achieves the so-called "free" by decreasing transmit power instead of introducing additional privacy-preserving mechanisms at devices. In this paper, we analyze leakage over multiple-input multiple-output (MIMO) fading channel. We show that multiple-antenna server amplifies leakage. Consequently, relying solely is inefficient meet high requirements, particularly when receive antenna array large. calls for joint optimization algorithm device-side mechanism and receiving protocol achieve better privacy-learning tradeoff. Numerical results validate our analysis highlight impact size

Language: Английский

Citations

1