Graph neural networks

Knows graph = nodes + edges; can sketch a social network.

Writes GCN layer h' = σ(A · h · W); implements on Cora.

Uses PyG / DGL; picks GAT / GraphSAGE / GIN by data; handles heterogeneous graphs.

Analyses over-smoothing, expressivity (WL test), positional encodings.

Contributes to graph transformers, equivariant GNNs, subgraph-level learning.