FLamby: Datasets and Benchmarks for Cross-Silo Federated Learning in Realistic Healthcare Settings
Jean Ogier Du Terrail
(1)
,
Samy-Safwan Ayed
(2)
,
Edwige Cyffers
(3)
,
Felix Grimberg
(4)
,
Chaoyang He
(5)
,
Regis Loeb
(1)
,
Paul Mangold
(3)
,
Tanguy Marchand
(1)
,
Othmane Marfoq
(6)
,
Erum Mushtaq
(7)
,
Boris Muzellec
(1)
,
Constantin Philippenko
(8)
,
Santiago Silva
(2)
,
Maria Teleńczuk
(1)
,
Shadi Albarqouni
(9, 10)
,
Salman Avestimehr
(7, 5)
,
Aurélien Bellet
(3)
,
Aymeric Dieuleveut
(8)
,
Martin Jaggi
(4)
,
Sai Praneeth Karimireddy
(11)
,
Marco Lorenzi
(12)
,
Giovanni Neglia
(13)
,
Marc Tommasi
(3)
,
Mathieu Andreux
(1)
1
Owkin France
2 UniCA - Université Côte d'Azur
3 MAGNET - Machine Learning in Information Networks
4 EPFL - Ecole Polytechnique Fédérale de Lausanne
5 FedML - FedML, Inc
6 IAC - Institut Agronomique Néo-Calédonien
7 USC - University of Southern California
8 CMAP - Centre de Mathématiques Appliquées de l'Ecole polytechnique
9 University Hospital Bonn
10 Helmholtz Munich, Deutsches Forschungszentrum für Gesundheit und Umwelt (GmbH)
11 UC Berkeley - University of California [Berkeley]
12 EPIONE - E-Patient : Images, données & mOdèles pour la médeciNe numériquE
13 NEO - Network Engineering and Operations
2 UniCA - Université Côte d'Azur
3 MAGNET - Machine Learning in Information Networks
4 EPFL - Ecole Polytechnique Fédérale de Lausanne
5 FedML - FedML, Inc
6 IAC - Institut Agronomique Néo-Calédonien
7 USC - University of Southern California
8 CMAP - Centre de Mathématiques Appliquées de l'Ecole polytechnique
9 University Hospital Bonn
10 Helmholtz Munich, Deutsches Forschungszentrum für Gesundheit und Umwelt (GmbH)
11 UC Berkeley - University of California [Berkeley]
12 EPIONE - E-Patient : Images, données & mOdèles pour la médeciNe numériquE
13 NEO - Network Engineering and Operations
Othmane Marfoq
- Fonction : Auteur
- PersonId : 739459
- IdHAL : othmane-marfoq
- ORCID : 0000-0002-0542-8925
Aurélien Bellet
- Fonction : Auteur
- PersonId : 9877
- IdHAL : aurelien-bellet
- ORCID : 0000-0003-3440-1251
- IdRef : 17653136X
Aymeric Dieuleveut
- Fonction : Auteur
- PersonId : 1109167
- IdHAL : aymeric-dieuleveut
- ORCID : 0009-0005-1848-1724
Marco Lorenzi
- Fonction : Auteur
- PersonId : 178572
- IdHAL : marco-lorenzi
- ORCID : 0000-0003-0521-2881
- IdRef : 168153335
Giovanni Neglia
- Fonction : Auteur
- PersonId : 1683
- IdHAL : giovanni-neglia
- ORCID : 0000-0001-8779-0620
- IdRef : 18310966X
Marc Tommasi
- Fonction : Auteur
- PersonId : 399
- IdHAL : marc-tommasi
- ORCID : 0000-0003-2838-4408
- IdRef : 121846385
Résumé
Federated Learning (FL) is a novel approach enabling several clients holding sensitive data to collaboratively train machine learning models, without centralizing data. The cross-silo FL setting corresponds to the case of few ($2$--$50$) reliable clients, each holding medium to large datasets, and is typically found in applications such as healthcare, finance, or industry. While previous works have proposed representative datasets for cross-device FL, few realistic healthcare cross-silo FL datasets exist, thereby slowing algorithmic research in this critical application. In this work, we propose a novel cross-silo dataset suite focused on healthcare, FLamby (Federated Learning AMple Benchmark of Your cross-silo strategies), to bridge the gap between theory and practice of cross-silo FL. FLamby encompasses 7 healthcare datasets with natural splits, covering multiple tasks, modalities, and data volumes, each accompanied with baseline training code. As an illustration, we additionally benchmark standard FL algorithms on all datasets. Our flexible and modular suite allows researchers to easily download datasets, reproduce results and re-use the different components for their research. FLamby is available at~\url{www.github.com/owkin/flamby}.
