From Graph Convolutional Network (GCN), we learned that combining local graph structure and node-level features yields good performance on node classification task. However, the way GCN aggregates is structure-dependent, which may hurt its generalizability.
One workaround is to simply average over all neighbor node features as in GraphSAGE. Graph Attention Network proposes an alternative way by weighting neighbor features with feature dependent and structure free normalization, in the style of attention.
The goal of this tutorial:
- Explain what is Graph Attention Network.
- Demonstrate how it can be implemented in DGL.
- Understand the attentions learnt.
- Introduce to...
Graph classification is an important problem with applications across many fields – bioinformatics, chemoinformatics, social network analysis, urban computing and cyber-security. Applying graph neural networks to this problem has been a popular approach recently (Ying et al., 2018, Cangea et al., 2018, Knyazev et al., 2018, Bianchi et al., 2019, Liao et al., 2019, Gao et al., 2019).
This tutorial is a demonstration for
- batching multiple graphs of variable size and shape with DGL
- training a graph neural network for a simple graph classification task
Simple Graph Classification Task
In this...Read more