Graph Neural Networks

Best for: Graph-structured data

How it works

A GNN updates each node’s feature vector by combining its own previous state with an aggregation of its neighbours, $h_v^{(k)}=\phi\bigl(W h_v^{(k-1)},\,\square_{u\in\mathcal{N}(v)}\psi(h_u^{(k-1)})\bigr)$, where $\square$ is a permutation-invariant aggregator (sum, mean or max). A Graph Convolutional Network uses the specific spectral update $H^{(k)}=\sigma(\tilde D^{-1/2}\tilde A\tilde D^{-1/2}H^{(k-1)}W)$ with the normalised adjacency, while GAT learns attention weights over neighbours. Stacking $K$ layers propagates information across $K$-hop neighbourhoods, and node- or graph-level readouts produce the final predictions.

Common fields

Drug discovery · social networks · fraud rings · recommendations