Federated multi-task learning with cross-device heterogeneous task subsets

Zewei Xin, Qinya Li, Chaoyue Niu, Fan Wu, Guihai Chen

Abstract: Highlights•We propose cross-device task heterogeneity, a collaborative scenario for federated clients with heterogeneous task sets.•We propose FedPMT, enabling clients with diverse task sets to collaboratively train cloud models, with proven convergence.•We use different model architectures locally and in the cloud for their tasks, enabling collaboration through alignment.•We propose task attenuation factors to promote task collaboration, enabling cloud models to converge to shared optima.•Extensive experiments verify the effectiveness of our method in various heterogeneous task set scenarios.

External IDs:dblp:journals/jpdc/XinLNWC25