On the benefits of Personalized Federated Learning
Event details
| Date | 26.08.2024 |
| Hour | 11:00 › 13:00 |
| Speaker | Abdellah El Mrini |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Martin Jaggi
Thesis advisor: Prof. Rachid Guerraoui
Co-examiner: Prof. Nicolas Flammarion
Abstract
Personalization is a promising paradigm for solving the problem of heterogeneity in Distributed Learning. As byproducts, some works in the literature pointed to many other benefits of personalization including robustness, fairness, and privacy. We aim to establish theoretical guarantees to prove and quantify the benefits of personalization and characterize the emerging tradeoffs.
Background papers
1- Tian Li, Shengyuan Hu, Ahmad Beirami, and Virginia Smith. Ditto: Fair and robust federated learning through personalization, ICML 2021
https://proceedings.mlr.press/v139/li21h/li21h.pdf
2- Mathieu Even, Laurent Massoulié, and Kevin Scaman. On sample optimality in personalized collaborative and federated learning. Neurips 2022
https://proceedings.neurips.cc/paper_files/paper/2022/file/01cea7793f3c68af2e4989fc66bf8fb0-Paper-Conference.pdf
3- Alberto Bietti, Chen-Yu Wei, Miroslav Dudik, John Langford, and Zhiwei Steven Wu. Personalization improves privacy-accuracy tradeoffs in federated learning, ICML 2022
https://proceedings.mlr.press/v162/bietti22a/bietti22a.pdf
Exam president: Prof. Martin Jaggi
Thesis advisor: Prof. Rachid Guerraoui
Co-examiner: Prof. Nicolas Flammarion
Abstract
Personalization is a promising paradigm for solving the problem of heterogeneity in Distributed Learning. As byproducts, some works in the literature pointed to many other benefits of personalization including robustness, fairness, and privacy. We aim to establish theoretical guarantees to prove and quantify the benefits of personalization and characterize the emerging tradeoffs.
Background papers
1- Tian Li, Shengyuan Hu, Ahmad Beirami, and Virginia Smith. Ditto: Fair and robust federated learning through personalization, ICML 2021
https://proceedings.mlr.press/v139/li21h/li21h.pdf
2- Mathieu Even, Laurent Massoulié, and Kevin Scaman. On sample optimality in personalized collaborative and federated learning. Neurips 2022
https://proceedings.neurips.cc/paper_files/paper/2022/file/01cea7793f3c68af2e4989fc66bf8fb0-Paper-Conference.pdf
3- Alberto Bietti, Chen-Yu Wei, Miroslav Dudik, John Langford, and Zhiwei Steven Wu. Personalization improves privacy-accuracy tradeoffs in federated learning, ICML 2022
https://proceedings.mlr.press/v162/bietti22a/bietti22a.pdf
Practical information
- General public
- Free