Graph U-Nets

Hongyang Gao, Shuiwang Ji, ICML-2019


This paper proposes a U-Net like architecture for graphical data and tries pretty good performance on node classification and graph classification tasks. Also for this task, they develop a novel pooling and unpooling techniques for graphical data, which is essential to get wider perspective during classification process, just like in the case of simple image data.

Main contributions

Implementation Details

Our two cents