README.md

SimGRACE: A Simple Framework for Graph Contrastive Learning without Data Augmentation

PyTorch implementation for SimGRACE: A Simple Framework for Graph Contrastive Learning without Data Augmentation accepted by The Web Conference 2022 (WWW 2022).

Overview

In this repository, we provide the codes of SimGRACE to evaluate its performances in terms of generalizability (unsupervised & semi-supervised learning), transferability (transfer learning) and robustness (adversarial robustness).

Dataset download

• Semi-supervised learning & Unsupervised representation learning [TU Datasets] (social and biochemical graphs)
• Transfer learning [chem data] (2.5GB);[bio data] (2GB)
• Adversarial robustness [synthetic data]

Citation

@inproceedings{xia2022SimGRACE,
 author = {Xia, Jun and Wu, Lirong and Chen, Jintao and Hu, Bozhen and Li, Stan Z.},
 booktitle = {The Web Conference},
 title = {SimGRACE: A Simple Framework for Graph Contrastive Learning without Data Augmentation},
 year = {2022}
}

Useful resources for Pretrained Graphs Models (PGMs)

• The first comprehensive survey for PGMs: [A Survey of Pretraining on Graphs: Taxonomy, Methods, and Applications]
• [A curated list of must-read papers, open-source pretrained models and pretraining datasets.]

Reference

1. Graph Contrastive Learning Automated (ICML 2021)
2. Graph Contrastive Learning with Augmentations (NeurIPS 2020)
3. Strategies for Pre-training Graph Neural Networks (ICLR 2020)
4. Adversarial Attack on Graph Structured Data (ICML 2018)