GCC: Graph Contrastive Coding for Graph Neural Network Pre-Training

Explore the innovative Graph Contrastive Coding (GCC) framework for self-supervised graph neural network pre-training in this 41-minute conference talk by Jiezhong Qiu at ICBS2024. Delve into the world of graph representation learning and its applications in real-world problems, including node classification, similarity search, and graph classification. Discover how GCC addresses the limitations of domain-specific models by capturing universal network topological properties across multiple networks. Learn about the subgraph instance discrimination pre-training task and the use of contrastive learning to develop intrinsic and transferable structural representations. Examine the extensive experiments conducted on three graph learning tasks and ten graph datasets, showcasing GCC's competitive or superior performance compared to task-specific and trained-from-scratch counterparts. Gain insights into the potential of the pre-training and fine-tuning paradigm for advancing graph representation learning.

Jiezhong Qiu: GCC: Graph Contrastive Coding for Graph Neural Network Pre-Training #ICBS2024