Do Gradient Inversion Attacks Make Federated Learning Unsafe?
IEEE Transactions on Medical Imaging · 2023-07 · 42 (7) · 2044-2056
DOI: 10.1109/tmi.2023.3239391 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Medical Imaging, 2023-07. Verfasst von Ali Hatamizadeh und 10 weiteren Autor:innen. Hauptthemen laut MEDLINE: Neural Networks, Computer, Supervised Machine Learning. Die Arbeit wurde bisher 126-mal zitiert (Crossref) und nennt 56 Literaturangaben.
Abstract (Originalsprache)
Federated learning (FL) allows the collaborative training of AI models without needing to share raw data. This capability makes it especially interesting for healthcare applications where patient and data privacy is of utmost concern. However, recent works on the inversion of deep neural networks from model gradients raised concerns about the security of FL in preventing the leakage of training data. In this work, we show that these attacks presented in the literature are impractical in FL use-cases where the clients' training involves updating the Batch Normalization (BN) statistics and provide a new baseline attack that works for such scenarios. Furthermore, we present new ways to measure and visualize potential data leakage in FL. Our work is a step towards establishing reproducible methods of measuring data leakage in FL and could help determine the optimal tradeoffs between privacy-preserving techniques, such as differential privacy, and model accuracy based on quantifiable metrics.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Zitieren
APA
Hatamizadeh, A., Yin, H., Molchanov, P., Myronenko, A., Li, W., Dogra, P., Feng, A., Flores, M. G., Kautz, J., Xu, D., & Roth, H. R. (2023). Do Gradient Inversion Attacks Make Federated Learning Unsafe?. IEEE Transactions on Medical Imaging, 42(7), 2044-2056. https://doi.org/10.1109/tmi.2023.3239391
Vancouver
Hatamizadeh A, Yin H, Molchanov P, Myronenko A, Li W, Dogra P, et al. Do Gradient Inversion Attacks Make Federated Learning Unsafe?. IEEE Transactions on Medical Imaging. 2023;42(7):2044-2056. doi:10.1109/tmi.2023.3239391
BibTeX
@article{Hatamizadeh2023,
author = {Hatamizadeh, Ali and Yin, Hongxu and Molchanov, Pavlo and Myronenko, Andriy and Li, Wenqi and Dogra, Prerna and Feng, Andrew and Flores, Mona G and Kautz, Jan and Xu, Daguang and Roth, Holger R},
title = {Do Gradient Inversion Attacks Make Federated Learning Unsafe?},
journal = {IEEE Transactions on Medical Imaging},
year = {2023},
volume = {42},
number = {7},
pages = {2044--2056},
doi = {10.1109/tmi.2023.3239391}
}
EUVIMED