Abstract:Federated Graph Learning (FGL) has emerged as a promising way to learn high-quality representations from distributed graph data with privacy preservation. Despite considerable efforts have been made for FGL under either cross-device or cross-silo paradigm, how to effectively capture graph knowledge in a more complicated cross-silo cross-device environment remains an under-explored problem. However, this task is challenging because of the inherent hierarchy and heterogeneity of decentralized clients, diversified privacy constraints in different clients, and the cross-client graph integrity requirement. To this end, in this paper, we propose a Hierarchical Federated Graph Learning (HiFGL) framework for cross-silo cross-device FGL. Specifically, we devise a unified hierarchical architecture to safeguard federated GNN training on heterogeneous clients while ensuring graph integrity. Moreover, we propose a Secret Message Passing (SecMP) scheme to shield unauthorized access to subgraph-level and node-level sensitive information simultaneously. Theoretical analysis proves that HiFGL achieves multi-level privacy preservation with complexity guarantees. Extensive experiments on real-world datasets validate the superiority of the proposed framework against several baselines. Furthermore, HiFGL's versatile nature allows for its application in either solely cross-silo or cross-device settings, further broadening its utility in real-world FGL applications.

What problem does this paper attempt to address?

The paper aims to address the challenges in cross-silo and cross-device Federated Graph Learning (FGL). Specifically: 1. **Multi-level Privacy Protection**: The paper proposes a Hierarchical Federated Graph Learning (HiFGL) framework to simultaneously protect subgraph-level and node-level privacy information, preventing information leakage to unauthorized participants. 2. **Challenges of Heterogeneous Clients**: The research addresses the technical difficulties arising from the inherent hierarchical structure and heterogeneity among different clients in cross-silo and cross-device environments, ensuring the effectiveness and integrity of the federated graph learning process. 3. **Diverse Privacy Constraints**: Flexible privacy protection strategies are proposed for different types of clients, considering subgraph-level privacy needs among institutions and node-level privacy protection on user devices. 4. **Integrity of Cross-client Graph Data**: While ensuring privacy, the integrity of cross-client graph data is maintained, avoiding the sacrifice of model performance due to privacy protection. Through the above methods, HiFGL can not only work effectively in complex cross-silo and cross-device scenarios but also be applied to single cross-silo or cross-device settings, demonstrating broad practical application potential.

HiFGL: A Hierarchical Framework for Cross-silo Cross-device Federated Graph Learning