On the Global Self-attention Mechanism for Graph Convolutional Networks

[tatsuropfgt]

On the Global Self-attention Mechanism for Graph Convolutional Networks [Wang+, 20]

Abstract

• Apply Global self-attention (GSA) to GCNs
• GSA allows GCNs to capture feature-based vertex relations regardless of edge connections

Background

• GSA mechanism on CNNs has achieved remarkable success
• CNNs and GCNs are similar, so they tried to apply GSA to GCNs

Detail

• experimented on node and graph classification tasks
• outperform multiple carefully designed advanced methods
• the impacts of overfitting and over-smoothing, two issues closely connected to graph edge structure
  • GSA can mitigate both issues
  • in section4, a theoretical analysis of these issues is discussed

[tatsuropfgt]