Understanding Pooling in Graph Neural Networks

  • https://arxiv.org/pdf/2110.05292.pdf
  • graph pooling 的综述,并对比了不同方法的效果
  • 各种 pooling 方法可以看作是三个函数的组合,selection, reduction, and connection (SRC). The selection function groups the nodes of the input graph into subsets called supernodes; then, the reduction function aggregates each supernode to form an output node and its attributes; finally, the connection function links the reduced nodes with (possibly attributed) edges and outputs the pooled graph.