Teaching Old Labels New Tricks In Heterogeneous Graphs

In a new research paper, Google Research Graph Mining Team introduces a new approach to heterogeneous graph neural networks (HGNNs) that tackles the label imbalance issue between different node types. This new approach, called “Zero-shot Transfer Learning within a Heterogeneous Graph via Knowledge Transfer”, allows HGNNs to be used in real world heterogeneous graphs, making them more broadly applicable.