Abstract: Federated Graph Learning (FGL) combines the privacy-preserving capabilities of federated learning (FL) with the strong graph modeling capability of Graph Neural Networks (GNNs). Current research addresses subgraph-FL only from the structural perspective, neglecting the propagation of graph signals on spatial and spectral domains of the structure. From a spatial perspective, subgraph-FL introduces edge disconnections between clients, leading to disruptions in label signals and a degradation in the class knowledge of the global GNN. From a spectral perspective, spectral heterogeneity causes inconsistencies in signal frequencies across subgraphs, which makes local GNNs overfit the local signal propagation schemes. As a result, spectral client drifts occur, undermining global generalizability. To tackle the challenges, we propose a global knowledge repository to mitigate label signal disruption and a frequency alignment to address spectral client drifts. The combination of $\textbf{S}$patial and $\textbf{S}$pectral strategies forms our framework $S^2$FGL. Extensive experiments on multiple datasets demonstrate the superiority of $S^2$FGL. The code is available at https://github.com/Wonder7racer/S2FGL.git.

Lay Summary:

In many real-world applications like fraud detection or social networks, data is distributed across different sources and cannot be shared due to privacy concerns. Federated Graph Learning (FGL) is a promising solution that trains models across multiple data holders without sharing their data. However, current FGL methods often struggle when the data is fragmented, leading to broken connections and inconsistent signals.Our paper introduces S2FGL, a new method that improves how graph neural networks learn from distributed data. It does this by addressing two key problems: the loss of useful label signals and differences in data frequency patterns across clients. S2FGL uses a shared knowledge base and frequency alignment to help each client learn more general and robust representations.This work provides a more reliable and effective way to train graph models in privacy-sensitive environments, with benefits for fields such as healthcare, finance, and social media analysis.