深度学习(6)：GNN

Graph Neural Network

Why do we need GNN?

classification：train a classifier

The data may have underlying structure and relationship\

• How do we utilize the structures and relationship to help our model?]
• What if the graph is larger, more than 20?
• What if we do not have the all the labels?
  

unlabeled Node >> labeled node

A node can learn the structure from its neighbors, but how?

How to embed node into feature space using convolution?

• solution 1: Generalize the concept of convolution to graph >> Spatial-based convolution
• solution 2: Back to the definition of convolution in signal processing>> Spectral-based GNN

GNN Roadmap

Representation learning: 学到一些比较好的features

Spatial-based GNN

1、NN4G

2、DCNN (Diffusion-Convolution Neural Network)

• 对于第0层的hidden layer，把距离v3点距离为1的node相加去平均乘一个权值
• 对于第1层的hidden layer，把距离node3点距离为2的node相加取平均乘权值。
  

3、DGC(Diffusion Graph Convolution)

4、MoNET （Mixture Model Networks）

• Define a measure on node “distances”
• Use weighted sum(mean) instead of simply summing up neighbor features

5、GraphSAGE

• Aggregation: mean, max-pooling, LSTM