Robust Optimization as Data Augmentation for Large-scale Graphs

This is the official repo for the paper Robust Optimization as Data Augmentation for Large-scale Graphs, accepted at CVPR2022.

TL;DR: FLAG augments node features to generalize GNNs on both node and graph classification tasks.

Highlights

• Simple, adding just a dozen lines of code
• General, applicable to any GNN baseline
• Versatile, working on both node and graph classification tasks
• Scalable, minimum memory overhead, working on the original infrastructure

Experiments

To reproduce experimental results for DeeperGCN, visit here.

Other baselines including GCN, GraphSAGE, GAT, GIN, MLP, etc. are available here.

To view the empirical performance of FLAG, please visit the Open Graph Benchmark Node and Graph classification leaderboards.

Requirements

• ogb>=1.2.3
• torch-geometric>=1.6.1
• torch>=1.5.0

Citing FLAG

If you find FLAG useful, please cite our paper.

@misc{https://doi.org/10.48550/arxiv.2010.09891,
  doi = {10.48550/ARXIV.2010.09891},
  url = {https://arxiv.org/abs/2010.09891},
  author = {Kong, Kezhi and Li, Guohao and Ding, Mucong and Wu, Zuxuan and Zhu, Chen and Ghanem, Bernard and Taylor, Gavin and Goldstein, Tom},
  keywords = {Machine Learning (cs.LG), Machine Learning (stat.ML), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {Robust Optimization as Data Augmentation for Large-scale Graphs},
  publisher = {arXiv},
  year = {2020},
  copyright = {arXiv.org perpetual, non-exclusive license}
}