CADET

Code for the paper "CADET: Calibrated Anomaly Detection for Mitigating Hardness Bias"

Demo Visualization

We have provided a juypter notebook demo.ipynb(link) for demo visualization.

Environment

• Python==3.6.12
• tensorflow==2.4.1
• tensorflow-gpu==2.2.0
• keras==2.4.3

More detail could be found in requirements.txt. (Some dependencies inside might not be necessary)

Some Instructions

• train.py: training the baseline models. The trained models will be saved under saved_models/ directory.
• save_error.py: getting and saving the errors or other information based on the trained models.
• eval.py: applying our method on the baselines.

Some example checkpoints (models or errors) are given via this link. Unzip it and place the subdirectories under the repo.

Citation

If you find this repo or our work useful for your research, please consider citing the paper

@inproceedings{deng2022cadet,
  title={CADET: Calibrated Anomaly Detection for Mitigating Hardness Bias},
  author={Deng, Ailin and Goodge, Adam and Ang, Lang Yi and Hooi, Bryan},
  booktitle={IJCAI},
  year={2022}
}