graph neural networks (GNN) are powerful models for many graph-structured
tasks. Existing models often assume that the complete structure of the graph is
available during training. In practice, however, graph-structured data is
usually formed in a streaming fashion so that learning a g