FedGraph: A New Paradigm for Federated Graph Learning
Keywords: Federated Learning, Anomaly Alignment, Network Alignment
Abstract: Federated learning is a distributed approach to training a global model over multiple clients without sharing their local data. In graph data, the data heterogeneity can correspond to subgraph structures and node features varying extremely different, and the task-specific knowledge isolation corresponds to exclusive schema on handing data for specific task in clients, e.g., anomaly user setting in Twitter is rather different from LinkedIn. Although most feder- ated graph learning approaches are employed to address the data heterogeneity challenge, we find that the task-specific knowledge isolation challenge has been overlooked. This task-specific knowl- edge isolation will prevent existing models into the federated graph learning framework. In this paper, we propose FedGraph: a new paradigm for federated graph learning. The key idea is to utilize the graph structure without private node features as structure knowl- edge bridging all task specific knowledge in clients. Our extensive experiments show that FedGraph significantly outperforms the other state-of-the-art federated learning algorithms on anomaly de- tection tasks. Two deep learning models and one existing anomaly subgraph detection model are transferred to FedGraph framework.
Primary Area: alignment, fairness, safety, privacy, and societal considerations
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9460
Loading