Classic Graph Convolutional Networks (GCNs) often learn node representation holistically, which would ignore the distinct impacts from different neighbors when aggregating their features to update a ...

Researchers have demonstrated a new training technique that significantly improves the accuracy of graph neural networks ...