Awesome
<div align="center"> <h1>0️⃣1️⃣ SpikeGCL (Spiking Graph Contrastive Learning)</h1> <h3>A Graph is Worth 1-bit Spikes: When Graph Contrastive Learning Meets Spiking Neural Networks</h3>Jintang Li<sup>1</sup>, Huizhe Zhang<sup>1</sup>, Ruofan Wu<sup>2</sup>, Zulun Zhu<sup>3</sup>, Baokun Wang<sup>2</sup>, Changhua Meng<sup>2</sup>, Zibin Zheng<sup>1</sup>, Liang Chen<sup>1</sup>
<sup>1</sup>Sun Yat-sen University, <sup>2</sup>Ant Group, <sup>3</sup>Nanyang Technological University
arXiv (arXiv:2305.19306), OpenReview (ICLR'24)
</div> <div align="center"> <img width="504" src="imgs/comparison.png"/> </div>
Environments
[!NOTE] Higher versions should be also compatible.
- numpy == 1.23.3
- torch == 1.8+cu111
- torch-cluster == 1.6.1
- torch_geometric == 2.3.0
- torch-scatter == 2.1.1
- torch-sparse == 0.6.17
- CUDA 11.1
- cuDNN 8.0.5
Model and Results
SpikeGCL adopts a simple GCL architecture and is comprised of a set of peer GNN encoders and a spiking neuron.
<div align="center"> <img src="imgs/spikegcl.png"/> </div>The following tables present the performance & efficiency results for standard node classification tasks on several graph benchmark datasets.
<div align="center"> <img src="imgs/tab1.png"/> <img src="imgs/tab2.png"/> </div>Reproduction
- Cora
python main.py --dataset Cora --threshold 5e-4 --outs 2 --T 64 --bn --epochs 5
- Citeseer
python main.py --dataset Citeseer --threshold 5e-3 --T 32 --bn --epochs 5
- Pubmed
python main.py --dataset Pubmed --threshold 5e-2 --bn --T 32 --epochs 50
- Computers
python main.py --dataset Computers --threshold 5e-2 --outs 32 --bn --T 25
- Photo
python main.py --dataset Photo --threshold 5e-2 --T 15 --bn --outs 8 --epochs 50
- CS
python main.py --dataset CS --threshold 5e-1 --outs 32 --T 60 --dropout 0. --bn
- Physics
python main.py --dataset Physics --T 25 --outs 16 --margin 1 --threshold 5e-2 --bn
- Ogbn-arXiv
python main.py --dataset ogbn-arxiv --T 30 --outs 1 --threshold 5e-2 --no_shuffle --bn --dropout 0.
- Ogbn-MAG
python main.py --dataset ogbn-mag --T 8 --outs 8 --hids 64 --threshold 5e-3 --no_shuffle --bn
Citation
If you find this repository useful in your research, please consider giving a star ⭐ and a citation
@inproceedings{spikegcl,
title={A Graph is Worth 1-bit Spikes: When Graph Contrastive Learning Meets Spiking Neural Networks},
author={Jintang Li and Huizhe Zhang and Ruofan Wu and Zulun Zhu and Baokun Wang and Changhua Meng and Zibin Zheng and Liang Chen},
booktitle={ICLR},
year={2024},
url={https://openreview.net/forum?id=LnLySuf1vp}
}