↓Skip to main content

Benefits of the Federation? Analyzing the Impact of Fair Federated Learning at the Client Level

·2 mins

TL;DR #

Benefits of the Federation? Analyzing the Impact of Fair Federated Learning at the Client Level asks whether joining a fair federated system benefits every client, rather than only improving a global metric.

Problem #

Federated learning brings together clients with different data distributions and potentially different fairness requirements. A fairness intervention that helps the federation overall may still reduce utility or fairness for a particular client.

What did we do? #

We studied the client-level effects of fair federated learning and compared methods including FedMinMax, PUFFLE, and reweighting strategies. The experiments evaluate both global outcomes and the performance and fairness experienced by individual clients.

Main findings #

Global fairness improvements do not automatically translate into benefits for every client. The results show that client heterogeneity can create trade-offs: a method can improve an aggregate objective while leaving some clients with worse local outcomes.

Resources #

Cite this paper #

@inproceedings{10.1145/3715275.3732152,
author = {Corbucci, Luca and Heilmann, Xenia and Cerrato, Mattia},
title = {Benefits of the Federation? Analyzing the Impact of Fair Federated Learning at the Client Level},
year = {2025},
isbn = {9798400714825},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3715275.3732152},
doi = {10.1145/3715275.3732152},
booktitle = {Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency},
pages = {2232–2248},
numpages = {17},
keywords = {Federated Learning, Fairness in Machine Learning, Bias, Algorithmic Fairness},
location = {Athens, Greece},
series = {FAccT '25}
}