> ML_ALGORITHM // GRAPH-CONVOLUTIONAL-NETWORKS-GCN_v1.0

Graph Convolutional Networks (GCN)

Foundational spatial-spectral graph neural network that aggregates feature information from immediate node neighborhoods via a normalized symmetric Laplacian.

Spectral & Spatial Graph Neural Networksgraph-relationalmoderate-posthocsmall (<1k)

Back to All Algorithms

Computational Complexity

Training Complexity:O(|E| * d + |V| * d^2)

Inference Complexity:O(|E| * d)

Hardware Profile

CPU Friendly:Yes

Requires GPU:Yes

Memory Footprint:moderate

Interpretability & Data

Interpretability Tier:moderate-posthoc

Training Data Needs:small (<1k)

Interpretability Assessment

Node embeddings are explicit linear combinations of normalized immediate neighbor features.

Suitable Tasks & Supported Modalities

Suitable Tasks:

node classificationgraph classificationlink prediction

Supported Modalities:

graphtabular

Implementing Libraries

torch-geometric

Deep Graph Library (DGL)DMLC / AWS / NYU · v2.2.1

View Spec

Foundational Literature

Semi-Supervised Classification with Graph Convolutional Networks (GCN)Thomas N. Kipf, Max Welling (2016) · International Conference on Learning Representations (ICLR)

Common Pitfalls & Warnings

Over-smoothing: stacking more than 3-4 GCN layers causes all node representations to converge to identical uniform vectors
Fails on heterophilous graphs where connected nodes have opposite labels