[GNN] graphic gnn:a gender Introduction (including video)

Catalog

GNN

Why

What

Tasks

 How

​ edit   The key  

MP

​ edit  GCN

​ edit  GAT

Reference resources

---

GNN

Why

What

 

 

Tasks

At present, the most common is node classification and link prediction

Figure classification is common in proteins 、 Pharmaceutical field  

 How

features X

Adjacency matrix A（ Connection ）

  Express by local neighborhood

 

2 layer GNN Example

Calculation C Nodes represent

 

  Calculation D node Express

  The key  

  The core ： Permutation invariance

That is, exchange any node order anyway , Will get a unique output , Corresponding to unique graph

 

  Equivariant

That is, input change , The output changes accordingly

 

MP

 GCN

violet ： Aggregate neighborhood

Blue ：W news ,f Activate ,B Parameters

Orange ： In fact, what has a greater impact on the final output is the node's own information

 GAT

Reference resources

https://www.youtube.com/watch?v=xFMhLp52qKI

CNN Medium equivariant vs. invariant - You know