Normal Structure Regularisation for Open-set Supervised GAD (ICLR 2025)

This is the offical Pytorch Implementation of ICLR 2025 paper Open-Set Graph Anomaly Detection via Normal Structure Regularisation.

By Qizhou Wang, Guansong Pang, Mahsa Salehi, Xiaokun Xia, Christopher Leckie.

Requirements

Please see the env.yml file.

Installation

conda env create -f env.yml

Usage

Please downlaod the dataset and set the path in exp/config/mag_cs/dset.yaml before running the code.

Please use the following script to run the training code:

# bash <run_script_name> <mode> <meta_config_name>
bash run_scripts/mag_cs/run.sh run meta_mag_cs

📝 Citation

If you find this work useful in your research, please consider citing:

@inproceedings{wang2024nsreg,
  title={Open-Set Graph Anomaly Detection via Normal Structure Regularisation}, 
  author={Qizhou Wang and Guansong Pang and Mahsa Salehi and Xiaokun Xia and Christopher Leckie},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year={2025},
}

🧾 License

This repository is released under the Apache 2.0 license as found in the LICENSE file.

---

Note: This repository is under active development. Code and detailed documentation will be released shortly.