Abstract: Graph neural networks have emerged as a leading architecture for many graph-level tasks such as graph classification and graph generation with a notable improvement. Among these tasks, graph pooling is an essential component of graph neural network architectures for obtaining a holistic graph-level representation of the entire graph. Although a great variety of methods have been proposed in this promising and fast-developing research field, to the best of our knowledge, little effort has been made to systematically summarize these methods. To set the stage for the development of future works, in this paper, we attempt to fill this gap by providing a broad review of recent methods on graph pooling. Specifically, 1) we first propose a taxonomy of existing graph pooling methods and provide a mathematical summary for each category; 2) next, we provide an overview of the libraries related to graph pooling, including the commonly used datasets, model architectures for downstream tasks, and open-source implementations; 3) then, we further outline in brief the applications that incorporate the idea of graph pooling in a number of domains; 4) and finally, we discuss some critical challenges faced by the current studies and share our insights on potential directions for improving graph pooling in the future.

Graph Pooling for Graph Neural Networks: Progress, Challenges, and Opportunities

A curated list of papers on graph pooling (More than 130 papers reviewed).

We provide a taxonomy of existing papers as shown in the above figure.

Papers in each category are sorted by their uploaded dates in descending order.

If you find this repository useful for your research, please consider citing our paper [arXiv]

@article{liu2022graph,
    title={Graph Pooling for Graph Neural Networks: Progress, Challenges, and Opportunities},
    author={Chuang Liu and Yibing Zhan and Chang Li and Bo Du and Jia Wu and Wenbin Hu and Tongliang Liu and Dacheng Tao},
    year={2022},
    journal={arXiv preprint arXiv:2204.07321},
}

• Analysis Papers

• Flat Pooling

• Hierarchical Pooling

  • Node Clustering Pooling
  • Node Drop Pooling
  • Other Pooling

• Applications

Analysis_Papers

1. Rethinking pooling in graph neural networks, NIPS 2020.

2. Understanding Pooling in Graph Neural Networks, ArXiv 2021.

3. A Fair Comparison of Graph Neural Networks for Graph Classification, ICLR 2020.

4. Graph Signal Processing and Deep Learning, IEEE Signal Processing Magazine 2020.

Flat_Pooling

Hierarchical_Pooling

Node_Clustering_Pooling

Node_Drop_Pooling

Other_Pooling

Applications

Download Source Code

Paper Preview