Semi-Supervised Classification with Graph Convolutional Networks (GCN)¶
Authors: Thomas N. Kipf, Max Welling Year: 2017 (arXiv 2016) Citations: 20,000+
Summary¶
Graph convolutional networks apply convolutions to graph-structured data. Spectral approximations enable efficient semi-supervised learning on graphs.
Key Concepts¶
- Graph Neural Networks: Extend CNNs to non-Euclidean data
- Spectral Convolutions: Convolutions via spectral theory
- Localization: Chebyshev polynomials approximate filters
- Semi-Supervised Learning: Learn from few labeled examples
- Neighborhood Aggregation: Aggregates from neighbors
- Scalability: Efficient computation on large graphs
Impact¶
- 20,000+ citations
- Foundation of modern graph neural networks
- Enabled learning on non-Euclidean data
- Influenced GraphSAGE, GAT, subsequent GNNs
- Applications in social, citation, knowledge graphs
- Opened geometric deep learning research
- Practical applications in node classification
Related Papers¶
- Spectral Networks (Bruna et al., 2013)
- GraphSAGE (Hamilton et al., 2017)
- Graph Attention Networks (Veličković et al., 2017)