Abstract:In the era of big data, managing evolving graph data poses substantial challenges due to storage costs and privacy issues. Training graph neural networks (GNNs) on such evolving data usually causes catastrophic forgetting, impairing performance on earlier tasks. Despite existing continual graph learning (CGL) methods mitigating this to some extent, they predominantly operate in centralized architectures and overlook the potential of distributed graph databases to harness collective intelligence for enhanced performance optimization. To address these challenges, we present a pioneering study on Federated Continual Graph Learning (FCGL), which adapts GNNs to multiple evolving graphs within decentralized settings while adhering to storage and privacy constraints. Our work begins with a comprehensive empirical analysis of FCGL, assessing its data characteristics, feasibility, and effectiveness, and reveals two principal challenges: local graph forgetting (LGF), where local GNNs forget prior knowledge when adapting to new tasks, and global expertise conflict (GEC), where the global GNN exhibits sub-optimal performance in both adapting to new tasks and retaining old ones, arising from inconsistent client expertise during server-side parameter aggregation. To tackle these, we propose the POWER framework, which mitigates LGF by preserving and replaying experience nodes with maximum local-global coverage at each client and addresses GEC by using a pseudo prototype reconstruction strategy and trajectory-aware knowledge transfer at the central server. Extensive evaluations across multiple graph datasets demonstrate POWER's superior performance over straightforward federated extensions of the centralized CGL algorithms and vision-focused federated continual learning algorithms. Our code is available at <a class="link-external link-https" href="https://github.com/zyl24/FCGL_POWER" rel="external noopener nofollow">this https URL</a>.

Toward Robust and Generalizable Federated Graph Neural Networks for Decentralized Spatial-Temporal Data Modeling

Decentralized Subgraph Learning for Spatial-Temporal Data Modeling

Federated Graph Semantic and Structural Learning

Graph Federated Learning Based on the Decentralized Framework

Cross-Node Federated Graph Neural Network for Spatio-Temporal Data Modeling

A Graph Neural Network Based Decentralized Learning Scheme

An Adaptive Federated Relevance Framework for Spatial Temporal Graph Learning

Enhancing Federated Graph Learning via Adaptive Fusion of Structural and Node Characteristics

Federated Graph Learning with Graphless Clients

Towards Federated Learning of Deep Graph Neural Networks

Towards Communication-Efficient Federated Graph Learning: an Adaptive Client Selection Perspective

Federated Graph Learning under Domain Shift with Generalizable Prototypes

Federated Temporal Graph Clustering

Decoupled Subgraph Federated Learning

FedSG: A Personalized Subgraph Federated Learning Framework on Multiple Non-IID Graphs

Tackling the Local Bias in Federated Graph Learning

A New Paradigm for Federated Structure Non-IID Subgraph Learning

Graph-regularized federated learning with shareable side information

Globally Consistent Federated Graph Autoencoder for Non-IID Graphs.

Federated Stochastic Configuration Networks for Distributed Data Analytics

Federated Continual Graph Learning