ElasticGNN

This repository includes the official implementation of ElasticGNN in the paper "Elastic Graph Neural Networks" [ICML 2021].

Xiaorui Liu, Wei Jin, Jiliang Tang at el. Elastic Graph Neural Networks.

Abstract

While many existing graph neural networks (GNNs) have been proven to perform L2-based graph smoothing that enforces smoothness globally, in this work we aim to further enhance the local smoothness adaptivity of GNNs via L1-based graph smoothing. As a result, we introduce a family of GNNs (Elastic GNNs) based on L1 and L2-based graph smoothing. In particular, we propose a novel and general message passing scheme into GNNs. This message passing algorithm is not only friendly to back-propagation training but also achieves the desired smoothing properties with a theoretical convergence guarantee. Experiments on semi-supervised learning tasks demonstrate that the proposed Elastic GNNs obtain better adaptivity on benchmark datasets and are significantly more robust to graph adversarial attacks.

Reference

Please cite our paper if you find the paper or code to be useful. Thank you!

@InProceedings{liu2021elastic,
  title = 	 {Elastic Graph Neural Networks},
  author =       {Liu, Xiaorui and Jin, Wei and Ma, Yao and Li, Yaxin and Liu Hua and Wang, Yiqi and Yan, Ming and Tang, Jiliang},
  booktitle = 	 {Proceedings of the 38th International Conference on Machine Learning},
  year = 	 {2021},
  series = 	 {Proceedings of Machine Learning Research},
  publisher =    {PMLR},
}

Requirement

• PyTorch
• PyTorch Geometric
• ogb (Evaluator)
• DeepRobust (optional)

Examples

Normal setting:

$ cd code
$ python3 main.py --dataset Cora --random_splits 10 --runs 1 --lr 0.01 --dropout 0.8  --weight_decay 0.0005 --K 10 --lambda1 3 --lambda2 3

Robustness setting

$ cd code
$ python3 main.py --dataset Cora-adv --random_splits 10 --runs 1 --lr 0.01 --K 10 --lambda1 9 --lambda2 3 --weight_decay 0.0005 --hidden 16 --normalize_features False --ptb_rate 0.1