Random Walk Neural Network (PyTorch)

Revisiting Random Walks for Learning on Graphs
Jinwoo Kim, Olga Zaghen*, Ayhan Suleymanzade*, Youngmin Ryou, Seunghoon Hong (* equal contribution)
ICML 2024 Workshop on Geometry-grounded Representation Learning and Generative Modeling

Updates

Sep 28, 2024

• Released the code for random walks and their records, and DeBERTa experiments.

Setup

Using Dockerfile (recommended)

git clone https://github.com/jw9730/random-walk.git /random-walk
cd random-walk
docker build --no-cache --tag rw:latest .
docker run -it --gpus all --ipc host --name rw -v /home:/home rw:latest bash
# upon completion, you should be at /rw inside the container

Using pip

git clone https://github.com/jw9730/random-walk.git /random-walk
cd random-walk
bash install.sh

Running Experiments

To try out random walks and their records, see the examples in the following files:

python3 test_walk_statistics.py
python3 test_walk_records.py

We will update the instructions for DeBERTa and llama 3 experiments soon.

Trained Models

We will release the trained model checkpoints for DeBERTa experiments soon.

References

Our implementation is based on code from the following repositories:

• kerighan/graph-walker for random walks
• ELENE for graph separation experiments
• Homomorphism Expressivity for substructure counting experiments

Citation

If you find our work useful, please consider citing it:

@article{kim2024revisiting,
  author    = {Jinwoo Kim and Olga Zaghen and Ayhan Suleymanzade and Youngmin Ryou and Seunghoon Hong},
  title     = {Revisiting Random Walks for Learning on Graphs},
  journal   = {arXiv},
  volume    = {abs/2407.01214},
  year      = {2024},
  url       = {https://arxiv.org/abs/2407.01214}
}