DIGNNs

This repository provides a reference implementation of DIGNN as described in the paper "Implicit Graph Neural Diffusion Based on Constrained Dirichlet Energy Minimization" which has been presented at NeurIPS 2023 New Frontiers in Graph Learning Workshop.

Requirements

• Install PyTorch >= 1.7.0
• Install PyTorch Geometric >= 1.7.0

Run Experiments

We provide some examples for running experiments for different tasks on different datasets:

Node classification

cd nodeclassification

For chameleon and squirrel datasets,

python main.py --input chameleon --model Neural --mu 2.2 --preprocess adj --max_iter 10 --dropout 0.5 --lr 0.01 --weight_decay 0

For PPI dataset,

python main_ppi.py --model Neural --dropout 0.1 --epoch 1000 --num_hid 512 --lr 0.01 --mu 2 --weight_decay 0 --max_iter 10

Graph classification

cd graphclassification

python main.py --input MUTAG --model Neural --mu 1.25 --max_iter 20 --num_hid 128 --lr 0.001 --weight_decay 0 --epochs 1000 

Citing

If you find DIGNN useful in your research, please cite our paper:

@article{DBLP:journals/corr/abs-2308-03306,
  author       = {Guoji Fu and
                  Mohammed Haroon Dupty and
                  Yanfei Dong and
                  Lee Wee Sun},
  title        = {Implicit Graph Neural Diffusion Based on Constrained Dirichlet Energy
                  Minimization},
  journal      = {CoRR},
  volume       = {abs/2308.03306},
  year         = {2023},
  url          = {https://doi.org/10.48550/arXiv.2308.03306},
  doi          = {10.48550/ARXIV.2308.03306},
  eprinttype    = {arXiv},
  eprint       = {2308.03306},
  timestamp    = {Mon, 21 Aug 2023 17:38:10 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2308-03306.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}