This repository contains the code for our ICLR 2024 accepted Spotlight paper, Unleashing the Potential of Fractional Calculus in Graph Neural Networks with FROND.
To install the required dependencies, refer to the environment.yaml file
To run our code, go to the /src folder.
python run_GNN_frac_all.py
--dataset Cora, Citeseer, Pubmed, CoauthorCS, CoauthorPhy, Computers, Photo
--function laplacian/ transformer
--block constant_frac/ att_frac
--method predictor/ predictor_corrector
--alpha_ode between (0,1] the value of beta in the paper
--time integration time
--step_size
FOR EXAMPLE:
run_GNN_frac_all.py --dataset Cora --function laplacian --block att_frac --cuda 1 --method predictor --epoch 400 --seed 123 --runtime 10 --decay 0.01 --dropout 0.2 --hidden_dim 256 --input_dropout 0.6 --alpha_ode 0.85 --time 40 --step_size 1.0 --lr 0.01
Our code is developed based on the following repo:
The FDE solver is from torchfde.
The graph neural ODE model is based on the GRAND, GraphCON, and GraphCDE framework.
If you find our work useful, please cite us as follows:
@INPROCEEDINGS{KanZhaDin:C24,
author = {Qiyu Kang and Kai Zhao and Qinxu Ding and Feng Ji and Xuhao Li and Wenfei Liang and Yang Song and Wee Peng Tay},
title={Unleashing the Potential of Fractional Calculus in Graph Neural Networks with {FROND}},
booktitle={Proc. International Conference on Learning Representations},
year={2024},
address = {Vienna, Austria},
note ={\textbf{spotlight}},
}