[2603.02233] Adaptive Personalized Federated Learning via Multi-task Averaging of Kernel Mean Embeddings

Computer Science > Machine Learning arXiv:2603.02233 (cs) [Submitted on 13 Feb 2026] Title:Adaptive Personalized Federated Learning via Multi-task Averaging of Kernel Mean Embeddings Authors:Jean-Baptiste Fermanian (PREMEDICAL), Batiste Le Bars (MAGNET, CRIStAL), Aurélien Bellet (PREMEDICAL) View a PDF of the paper titled Adaptive Personalized Federated Learning via Multi-task Averaging of Kernel Mean Embeddings, by Jean-Baptiste Fermanian (PREMEDICAL) and 3 other authors View PDF Abstract:Personalized Federated Learning (PFL) enables a collection of agents to collaboratively learn individual models without sharing raw data. We propose a new PFL approach in which each agent optimizes a weighted combination of all agents' empirical risks, with the weights learned from data rather than specified a priori. The novelty of our method lies in formulating the estimation of these collaborative weights as a kernel mean embedding estimation problem with multiple data sources, leveraging tools from multi-task averaging to capture statistical relationships between agents. This perspective yields a fully adaptive procedure that requires no prior knowledge of data heterogeneity and can automatically transition between global and local learning regimes. By recasting the objective as a high-dimensional mean estimation problem, we derive finite-sample guarantees on local excess risks for a broad class of distributions, explicitly quantifying the statistical gains of collaboration. To addre...