subgraph neural networks论文笔记

Subgraph Neural Network

• Graph-level representation: loss some finer local structure
• node-level: preserve local topological structure, detriment the big picture
  
  Problem Formulation:
  给出子图集合 s = { S 1 , S 2 . . . S N } s= \left\{ S_1, S_2...S_N \right\} s={ S1​,S2​...SN​}, SUBGNN给出产生d维子图向量表示的神经信息传递架构Es并进行分类

subgraph拓扑结构properties

Border neighborhood, as with nodes, is defined as the set of nodes within k hops
of any node in S.

SubGNN

Subgraph level message passing

• 从图中按通道选取子图作为Anchor patches，从而得到Ap, An, As
  对于每个channel，有sampling函数
  

  1. position channel:
     internal：sample函数对于每个子图只返回一个节点, anchor patches $A_{PA}$, shared across all components in S
     external：anchor patches