Distributed Graph Neural Networks

Graph neural networks (GNN) have made tremendous progress in recent years and have achieved state-of-the-art performance in diverse applications: recommender systems, anomaly detection, and social network analysis. GNNs use message-passing to aggregate information from neighborhoods to learn representations. As the real-world graphs are very large, it is essential to develop distributed GNN frameworks. In this tutorial, we explain the core components of the Distributed Deep Graph Library (DistDGL) and how to use it and extend it for research and developing new applications. We give examples of adding new partitioning, sampling, and personalization approaches to DistDGL based on recent papers. We also compare alternate frameworks and some open challenges and research problems.