Abstract

Graph neural networks have emerged as a powerful tool for learning on graphs and have attracted great attention in recent years. GNNs have been shown to be effective not only for several graph-related learning tasks, such as node classification, link prediction, and graph generation, but also in other domains, such as natural language processing, computer vision, and cybersecurity. However, despite the rapid growth in the field, scalability remains a critical issue in building GNN models for large graphs, as real-world graphs are continuously growing, often reaching sizes of billions of nodes and edges, or even a trillion edges. Therefore, it is desirable to develop efficient and scalable solutions to scale GNN training to very large graphs. In this talk, I will present SDT-GNN, a memory-efficient distributed GNN training framework for large graphs under limited computational resources. SDT-GNN is powered by a novel streaming-based graph partitioning algorithm named SPRING, which partitions large graphs effectively and efficiently. SDT-GNN has up to 95% less memory footprint than mainstream distributed GNN frameworks without sacrificing training speed and model accuracy. I will also present SpLPG, a communication-efficient distributed GNN training framework for link prediction that reduces the communication overhead by up to 80% while mostly preserving link prediction accuracy.

About the Speaker

Chul-Ho Lee is currently an Associate Professor in the Department of Computer Science at Texas State University, San Marcos, TX. Prior to joining Texas State, he was an Assistant Professor in the Department of Computer Engineering and Sciences at the Florida Institute of Technology, Melbourne, FL, and a Senior Research Engineer at Samsung Electronics DMC R&D Center, South Korea. He received his Ph.D. in Computer Engineering from North Carolina State University, Raleigh, NC. His recent research focuses on developing efficient algorithms and computational tools for large-scale graph mining and network analysis, and advancing machine-learning techniques for mobile, IoT and other applications. His research has been supported by funding agencies and industry partners including the National Science Foundation, NVIDIA, and SK Hynix America.