This is our PyTorch implementation for the paper:
Sihang Li, Xiang Wang*, An Zhang, Ying-Xin Wu, Xiangnan He and Tat-Seng Chua (2022). Let Invariant Rationale Discovery Inspire Graph Contrastive Learning, Paper in arXiv. In ICML'22, Baltimore, Maryland, USA, July 17-23, 2022.
Author: Sihang Li (sihang0520 at gmail.com)
Without supervision signals, Rationale-aware Graph Contrastive Learning (RGCL) uses a rationale generator to reveal salient features about graph instance-discrimination as the rationale, and then creates rationale-aware views for contrastive learning. This rationale-aware pre-training scheme endows the backbone model with the powerful representation ability, further facilitating the fine-tuning on downstream tasks.
If you want to use our codes and datasets in your research, please cite:
@inproceedings{RGCL,
author = {Sihang Li and
Xiang Wang and
An Zhang and
Xiangnan He and
Tat-Seng Chua},
title = {Let Invariant Rationale Discovery Inspire Graph Contrastive Learning},
booktitle = {{ICML}},
year = {2022}
}
- Transfer Learning on MoleculeNet datasets
- Semi-supervised learning on Superpixel MNIST dataset
- Unsupervised representation learning on TU datasets
Some issues might occur due to the version mismatch.
KeyError:'num_nodes'in unsupervised_TU: Shen-Lab/GraphCL#36, Shen-Lab/GraphCL#41AttributeError: 'Data' object has no attribute 'cat_dim'in transferLearning_MoleculeNet_PPI: Shen-Lab/GraphCL#13
The backbone implementation is reference to https://github.com/Shen-Lab/GraphCL.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent