Deeper Insights into Graph Convolutional Networks for Semi-Supervised Learning

Abstract

Many interesting problems in machine learning are being revisited with new deep learning tools. For graph-based semisupervised learning, a recent important development is graph convolutional networks (GCNs), which nicely integrate local vertex features and graph topology in the convolutional layers. Although the GCN model compares favorably with other state-of-the-art methods, its mechanisms are not clear and it still requires a considerable amount of labeled data for validation and model selection. In this paper, we develop deeper insights into the GCN model and address its fundamental limits. First, we show that the graph convolution of the GCN model is actually a special form of Laplacian smoothing, which is the key reason why GCNs work, but it also brings potential concerns of over-smoothing with many convolutional layers. Second, to overcome the limits of the GCN model with shallow architectures, we propose both co-training and self-training approaches to train GCNs. Our approaches significantly improve GCNs in learning with very few labels, and exempt them from requiring additional labels for validation. Extensive experiments on benchmarks have verified our theory and proposals.

AAAI-2018 Oral Presentation

Cited by in corpus (39)

• Simplifying Graph Convolutional Networks
• Spatial-Temporal Graph ODE Networks for Traffic Flow Forecasting
• Graph-Based Deep Learning for Medical Diagnosis and Analysis: Past, Present and Future
• A Note on Over-Smoothing for Graph Neural Networks
• Towards Deeper Graph Neural Networks with Differentiable Group Normalization
• TransGCN:Coupling Transformation Assumptions with Graph Convolutional Networks for Link Prediction
• When Does Self-Supervision Help Graph Convolutional Networks?
• Self-supervised Consensus Representation Learning for Attributed Graph
• Zero-shot Node Classification with Decomposed Graph Prototype Network
• Meta-GNN: On Few-shot Node Classification in Graph Meta-learning
• Understanding Negative Sampling in Graph Representation Learning
• Improving Graph Attention Networks with Large Margin-based Constraints
• ROD: Reception-aware Online Distillation for Sparse Graphs
• GLMNet: Graph Learning-Matching Networks for Feature Matching
• Gophormer: Ego-Graph Transformer for Node Classification
• Edge Representation Learning with Hypergraphs
• Learning to Drop: Robust Graph Neural Network via Topological Denoising
• DEMO-Net: Degree-specific Graph Neural Networks for Node and Graph Classification
• Revisiting Graph based Collaborative Filtering: A Linear Residual Graph Convolutional Network Approach
• Node Similarity Preserving Graph Convolutional Networks
• Signed Graph Diffusion Network
• RGAT: A Deeper Look into Syntactic Dependency Information for Coreference Resolution
• Dual ResGCN for Balanced Scene GraphGeneration
• A Hierarchy of Graph Neural Networks Based on Learnable Local Features
• Graph Ensemble Learning over Multiple Dependency Trees for Aspect-level Sentiment Classification
• Causal Discovery from Incomplete Data using An Encoder and Reinforcement Learning
• Beyond Homophily with Graph Echo State Networks
• Graph Neural Distance Metric Learning with Graph-Bert
• Higher-Order Implicit Fairing Networks for 3D Human Pose Estimation
• GraLSP: Graph Neural Networks with Local Structural Patterns
• Tensor Graph Convolutional Networks for Text Classification
• LEReg: Empower Graph Neural Networks with Local Energy Regularization
• Multi-Zone Unit for Recurrent Neural Networks
• Semantic Graph Convolutional Network for Implicit Discourse Relation Classification
• G5: A Universal GRAPH-BERT for Graph-to-Graph Transfer and Apocalypse Learning
• On Local Aggregation in Heterophilic Graphs
• Multiplex Bipartite Network Embedding using Dual Hypergraph Convolutional Networks
• Effective Eigendecomposition based Graph Adaptation for Heterophilic Networks
• Mutual Teaching for Graph Convolutional Networks