This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Zihan Tan, Suyuan Huang, Guancheng Wan, Wenke Huang, He Li, Mang Ye
Outline
This paper presents Federated Graph Learning (FGL), which combines the privacy-preserving capabilities of Federated Learning (FL) with the powerful graph modeling capabilities of Graph Neural Networks (GNNs). Existing research has focused on subgraph FL from a structural perspective, but has overlooked the propagation of graph signals in the spatial and spectral domains of the structure. From a spatial perspective, subgraph FL causes edge disconnection between clients, leading to label signal disruption and a degradation of the semantic knowledge of the global GNN. From a spectral perspective, spectral heterogeneity causes signal frequency mismatches across subgraphs, causing local GNNs to overfit to local signal propagation methods. This results in spectral client drift, which degrades global generalization performance. To address these issues, this paper proposes a global knowledge repository to mitigate the lack of semantic knowledge caused by label signal disruption. Furthermore, we design a frequency alignment method to address spectral client drift. We construct the S2FGL framework by combining spatial and spectral strategies. Extensive experiments on multiple datasets demonstrate the superiority of S2FGL. The code is available at https://github.com/Wonder7racer/S2FGL.git .
We clearly present the spatial and spectral problems that arise in subgraph FL and propose S2FGL, an effective framework to solve them.
◦
We experimentally demonstrate that the performance of federated graph learning can be improved by using a global knowledge repository and frequency sorting techniques.
◦
We improve reproducibility through open code and support further research by other researchers.
•
Limitations:
◦
The effectiveness of the proposed method may vary depending on the dataset and model used. Further performance evaluation in various environments is required.
◦
Further research is needed on the size and management of global knowledge repositories. As the size of the repositories increases, performance degradation may occur.
◦
Further research is needed to apply this method to real-world applications. For example, a more in-depth analysis of communication overhead and privacy guarantees may be necessary.
