FedHyperGraph: A layer-wise personalized federated learning with correlation graphs in hyperbolic space

The goal of Personalized Federated Learning (PFL) is to achieve model personalization in statistically heterogeneous scenarios. However, most existing approaches treat model parameters as a monolithic unit during aggregation. This coarse-grained strategy overlooks the latent correlations among parameters across different clients, which are crucial when clients share similar tasks but possess heterogeneous data. To address this limitation, we propose FedHyperGraph, a graph-guided aggregation framework for layer-wise PFL. FedHyperGraph captures layer-wise knowledge to construct latent correlation graphs in hyperbolic space, which effectively guide the personalized aggregation process. Extensive experiments demonstrate that FedHyperGraph significantly outperforms state-of-the-art baselines, achieving accuracy improvements of up to 42.6 %, 33.1 %, and 7.5 % on graph, computer vision (CV), and natural language processing (NLP) datasets, respectively. Furthermore, FedHyperGraph accelerates convergence by up to 52.6 % on CV tasks and demonstrates superior scalability across varying client scales.