GNN Graph Isomorphism方向探索

  • Post author:
  • Post category:其他





前言

本文主要讨论GNN在graph isomorphic方面的可行理论研究方向。




一、知识储备


  1. Weisfeiler-Lehman Test



二、各论文主体



1. ICLR2019:GNN有多牛

An interesting direction for future work is to go beyond neighborhood aggregation (or message passing) in order to pursue possibly even more powerful architectures for learning with graphs. To complete the picture, it would also be interesting to understand and improve the generalization properties of GNNs as well as better understand their optimization landscape.



2. ICLR2017:基于GNN的半监督学习

作者对Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering这个工作进行了简化,使之应用于graph节点的半监督分类问题,取得了不错的效果.



3. NIPS2017:适用于大规模网络的归纳式(inductive)学习方法



4. 1968:将图化简为规范形式



三、可行方向

现有手段还是根据不同的embedding得出不同构的结论,但是怎样直接根据GNN判断时候同构还没有很好的方法,可以在从方面着手



版权声明:本文为weixin_43851192原创文章,遵循 CC 4.0 BY-SA 版权协议,转载请附上原文出处链接和本声明。