On Addressing the Limitations of Graph Neural Networks

2023/06/22 by Luan, Sitao
#FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2306.12640

Abstract

This report gives a summary of two problems about graph convolutional networks (GCNs): over-smoothing and heterophily challenges, and outlines future directions to explore.

Related

• GPNet: Simplifying Graph Neural Networks via Multi-channel Geometric Polynomials
• What Do Graph Convolutional Neural Networks Learn?
• Graph Neural Networks: Architectures, Stability and Transferability
• DeeperGCN: All You Need to Train Deeper GCNs
• Graph Convolutional Networks: analysis, improvements and results