GCN (Graph Convolutional Network, Kipf & Welling 2017) propagates as
H(l+1)=σ(D~−1/2A~D~−1/2H(l)W(l)), where
A~=A+I adds self-loops,
D~ is its degree matrix,
W(l) is learnable, and
σ is an activation — essentially “degree-normalized weighted average of neighbor features + linear transform”. Spectral motivation: the graph convolution
x∗gθ=Ugθ(Λ)UTx requires
O(n2) eigen-decomposition; approximating
gθ(Λ)≈∑k=0KθkTk(Λ^) with Chebyshev polynomials and truncating at first order (
K=1), sharing the two parameters (
θ0=−θ1=θ) and renormalizing (
A~=A+I) yields exactly the formula above — each node aggregates its 1-hop neighbors, and stacking
L layers expands the receptive field to
L hops. The cost is
oversmoothing: after many layers node representations collapse toward the dominant eigenvectors and become indistinguishable (repeated low-pass filtering erases high-frequency information), so classic GCNs are typically 2–3 layers; mitigations include residual connections, JK-Net (jump connections with pooling), PairNorm, and layer normalization.