← ML Research Wiki / 1609.02907

SEMI-SUPERVISED CLASSIFICATION WITH GRAPH CONVOLUTIONAL NETWORKS

Thomas N Kipf [email protected] Canadian Institute for Advanced Research (CIFAR) University of Amsterdam University of Amsterdam, Max Welling [email protected] Canadian Institute for Advanced Research (CIFAR) University of Amsterdam University of Amsterdam (2016)

Paper Information
arXiv ID
1609.02907
Venue
International Conference on Learning Representations
Domain
Machine learning, graph neural networks, deep learning
SOTA Claim
Yes
Reproducibility
8/10
Contents

Abstract

We present a scalable approach for semi-supervised learning on graph-structured data that is based on an efficient variant of convolutional neural networks which operate directly on graphs. We motivate the choice of our convolutional architecture via a localized first-order approximation of spectral graph convolutions. Our model scales linearly in the number of graph edges and learns hidden layer representations that encode both local graph structure and features of nodes. In a number of experiments on citation networks and on a knowledge graph dataset we demonstrate that our approach outperforms related methods by a significant margin.

Summary

This paper presents a scalable approach for semi-supervised learning on graph-structured data through a variant of convolutional neural networks directly applied to graphs. The authors motivate their architecture with a localized first-order approximation of spectral graph convolutions, resulting in a model that scales linearly with graph edges, enhancing classification by encoding graph structure and node features. Key contributions include a layer-wise propagation rule for graph neural networks and an efficient model for semi-supervised node classification, demonstrated through experiments on citation networks and knowledge graphs. Experimental results showcase superior performance against existing methods in both accuracy and efficiency.

Methods

This paper employs the following methods:

  • Graph Convolutional Network (GCN)

Models Used

  • GCN

Datasets

The following datasets were used in this research:

  • Citeseer
  • Cora
  • Pubmed
  • NELL

Evaluation Metrics

  • Accuracy

Results

  • Outperforms related methods in classification accuracy and efficiency on citation networks and knowledge graphs.

Limitations

The authors identified the following limitations:

  • Memory requirements grow with dataset size requiring possible mini-batch SGD for large graphs.
  • Current framework does not support directed edges or edge features naturally.
  • Assumptions about locality and equivalence of self-connections may limit model applicability in certain datasets.

Technical Requirements

  • Number of GPUs: None specified
  • GPU Type: None specified

Keywords

semi-supervised learning graph convolutional networks spectral convolution node classification spectral graph theory

Papers Using Similar Methods

External Resources

Funding: SAP
References: 38
Influential Citations: 6498