ProFL: Performative Robust Optimal Federated Learning
IEEE Transactions on Signal Processing · 2026 · 74 · 4740-4756
DOI: 10.1109/tsp.2026.3742668 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Signal Processing, 2026. Verfasst von Xue Zheng und 4 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Gefördert von: NSF.
Abstract (Originalsprache)
Performative prediction is a framework that captures distribution shifts that occur during the training of machine learning models due to their deployment. As the trained model is used, the induced data distribution evolves, leading to deviations from the original data distribution. The impact of such model-induced distribution shifts in federated learning (FL) is increasingly likely to transpire in real-life use cases. A recently proposed approach extends performative prediction to FL with the resulting model converging to a performative stable point, which may be far from the performative optimal point. Earlier research in centralized settings has shown that the performative optimal point can be achieved under model-induced distribution shifts, but these approaches require the performative risk to be convex and the training data to be noiseless. These assumptions are often violated in realistic FL systems. This paper addresses these limitations and proposes Performative Robust Optimal Federated Learning (ProFL), an algorithm that targets the performative optimal point in FL while mitigating the effects of noisy and contaminated local data. We present a neighborhood-convergence analysis under the Polyak–Łojasiewicz condition, which applies to non-convex objectives. Extensive experiments on multiple datasets demonstrate the advantage of Robust Optimal Federated Learning over the state-of-the-art.
Förderung
- NSF
Zitieren
APA
Zheng, X., Xie, T., Tan, X., Zhang, X., & Yener, A. (2026). ProFL: Performative Robust Optimal Federated Learning. IEEE Transactions on Signal Processing, 74, 4740-4756. https://doi.org/10.1109/tsp.2026.3742668
Vancouver
Zheng X, Xie T, Tan X, Zhang X, Yener A. ProFL: Performative Robust Optimal Federated Learning. IEEE Transactions on Signal Processing. 2026;74:4740-4756. doi:10.1109/tsp.2026.3742668
BibTeX
@article{Zheng2026,
author = {Zheng, Xue and Xie, Tian and Tan, Xuwei and Zhang, Xueru and Yener, Aylin},
title = {ProFL: Performative Robust Optimal Federated Learning},
journal = {IEEE Transactions on Signal Processing},
year = {2026},
volume = {74},
pages = {4740--4756},
doi = {10.1109/tsp.2026.3742668}
}
EUVIMED