Answers are grounded in this paper and cite their sources.
Federated Learning (FL) is a privacy-preserving distributed machine learning paradigm, yet heterogeneous data often degrades the performance of shared global models. Clustered FL methods, which group clients with similar data distributions and train separate models, help mitigate this issue. However, existing clustered FL algorithms typically rely on local model-related information, leading to computational and communication overhead, and require pre-specifying the number of clusters, which is impractical in real-world scenarios. To overcome these limitations, this work explores an affinity propagation-based clustered FL framework, DACFL, which is directly aware of data discrepancies and infers cluster structures without prior knowledge. It incorporates a set of metrics to characterize feature distribution shift, label distribution shift, and their cooccurrence, aiming to improve the clustering process. Additionally, to protect label-related information during clustering, DACFL includes a novel scheme using fully homomorphic encryption to ensure secure use of label information. Experiments on public datasets suggest that the proposed approach performs favorably compared with several recent FL methods under heterogeneous settings.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: