Enhancing Trustworthiness of Graph Neural Networks with Rank-Based Conformal Training 🔥

Data download

All dataset can be downloaded here.

Run RCP-GNN

The hyper-parameters used to train the model is set as default in the optimal_param_set.pkl in the training files. Feel free to change them if needed.

Simply run bellow command to reproduce the results in the paper.

python main_smooth.py --model GCN \
                --dataset Cora_ML_CF \
                --device cuda:0 \
                --alpha 0.1\
                --conformal_score thrrank\
                --not_save_res\
                --interpolation higher\
                --num_runs 1\
                --conftr_calib_holdout\
                --conftr\
                --verbose\

Installation

We implement our code by TrochCP toolbox. See requirements.txt for the packages.

Reference

Please cite our work if you find it useful:

@misc{wang2025enhancingtrustworthinessgraphneural,
      title={Enhancing Trustworthiness of Graph Neural Networks with Rank-Based Conformal Training}, 
      author={Ting Wang and Zhixin Zhou and Rui Luo},
      year={2025},
      eprint={2501.02767},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2501.02767}, 
}