harishgovardhandamodar / ss-gcns

[ICML 2020] "When Does Self-Supervision Help Graph Convolutional Networks?" by Yuning You, Tianlong Chen, Zhangyang Wang, Yang Shen

When Does Self-Supervision Help Graph Convolutional Networks?

PyTorch implementation for When Does Self-Supervision Help Graph Convolutional Networks? [appendix]

Yuning You^*, Tianlong Chen^*, Zhangyang Wang, Yang Shen

In ICML 2020.

Overview

Properly designed multi-task self-supervision benefits GCNs in gaining more generalizability and robustness. In this repository we verify it through performing experiments on several GCN architectures with three designed self-supervised tasks: node clustering, graph partitioning and graph completion.

Dependencies

Please setup the environment following Section 3 (Setup Python environment for GPU) in this instruction, and then install the dependencies related to graph partitioning with the following commands:

sudo apt-get install libmetis-dev
pip install METIS==0.2a.4

Experiments

• GCN, GAT and GIN with self-supervision
• GMNN and GraphMix with self-supervision
• GCN with self-supervision in adversarial defense

Citation

If you use this code for you research, please cite our paper.

@article{you2020does,
  title={When Does Self-Supervision Help Graph Convolutional Networks?},
  author={You, Yuning and Chen, Tianlong and Wang, Zhangyang and Shen, Yang},
  journal={Proceedings of machine learning research},
  volume={119},
  pages={10871--10880},
  year={2020}
}