Contents

L17-huge GNN

Scaling up GNNs

/l17-huge-gnn/image.png

直接load全部nodes又不太可能【naive approach】4090 / A100带不动

neighbor sampling

/l17-huge-gnn/image-1.png 对hub node的思考 /l17-huge-gnn/image-2.png

/l17-huge-gnn/image-3.png see the paper

cluster-GCN

/l17-huge-gnn/image-4.png

/l17-huge-gnn/image-5.png

/l17-huge-gnn/image-6.png

advanced /l17-huge-gnn/image-7.png

Simplified GCN

舍弃了GCN的non-linearity,直接用linear layer

同质性?但是我想知道和glidar的区别? /l17-huge-gnn/image-8.png

/l17-huge-gnn/image-9.png