FedSGC: Federated Simple Graph Convolution for Node Classification
Published in International Workshop on Federated and Transfer Learning for Data Sparsity and Confidentiality in Conjunction with IJCAI 2021 (FTL-IJCAI-21), 2021
Graph Neural Networks (GNN) have developed rapidly and solved a wide range of graph-related tasks. One advantage of GNN over traditional neural networks is the utilization of neighbouring information. However, under the increasing concern in data privacy, accessing raw information from different parties may raise privacy concerns. To address this, federated learning, as a distributed learning mechanism, is proposed to train models with decentralized data owned by different data parties without sharing or leaking the raw data. In this work, we study the vertical and horizontal settings for federated learning on graph data. We propose FedSGC to train the Simple Graph Convolution model under three data split scenarios. We also demonstrate that the prediction performance of FedSGC is closely aligned with the non-federated model trained in centralized manner.