Data-Driven Laplacian-Constrained Personalized Federated Learning

Personalization addresses one of the defining challenges in federated learning: statistical heterogeneity among clients. We show that the dissimilarity between the ideal personalized models of two clients can be upper bounded by the dissimilarity of their distributions. Motivated by this, we propose...

Disgrifiad llawn

Manylion Llyfryddiaeth

Prif Awduron:	Erickson, Samuel, Johansson, Mikael
Fformat:	Erthygl
Cyhoeddwyd:	2025
Pynciau:	Machine Learning
Mynediad Ar-lein:	https://arxiv.org/abs/2505.07575 http://arxiv.org/abs/2505.07575
Tagiau:	Ychwanegu Tag Dim Tagiau, Byddwch y cyntaf i dagio'r cofnod hwn!