FGWMixup: Fused Gromov-Wasserstein Graph Mixup for Graph-level Classifications

This is the code for the paper: Fused Gromov-Wasserstein Graph Mixup for Graph-level Classifications, published in NeurIPS'23.

Paper link 🔗:

arXiv: https://arxiv.org/abs/2306.15963

OpenReview: https://openreview.net/forum?id=uqkUguNu40&noteId=0qcp06CFB6

Thanks for your interest in our work! If our work helps, please don't forget to cite us!🌟

@inproceedings{ma2023fused,
 author = {Ma, Xinyu and Chu, Xu and Wang, Yasha and Lin, Yang and Zhao, Junfeng and Ma, Liantao and Zhu, Wenwu},
 booktitle = {Advances in Neural Information Processing Systems},
 pages = {15252--15276},
 title = {Fused Gromov-Wasserstein Graph Mixup for Graph-level Classifications},
 url = {https://proceedings.neurips.cc/paper_files/paper/2023/file/3173c427cb4ed2d5eaab029c17f221ae-Paper-Conference.pdf},
 volume = {36},
 year = {2023}
}

File Structure

• ./src/: source codes

  gmixup_dgl.py: Main python file to run FGWMixup

  gromov_mixup.py: Conducting mixup of two samples

  FGW_barycenter.py: Calculating FGW barycenter and its accelerated version

  models_dgl.py: GNN architectures

  utils_dgl.py: Some utilities

• run_gmixup.sh: sh command to run FGWMixup

Requirements

Suggested Enviornments:

• Python 3.9
• PyTorch 1.11.0
• DGL 1.0.2
• POT 0.8.2