Sensitive-Channel-Pruning

This is the official repository of the following paper: "Achieving Fairness Through Channel Pruning for Dermatological Disease Diagnosis" in Proc. of Medical Image Computing and Computer Assisted Interventions (MICCAI), 2024 (early accept, acceptance rate 11%)

Environment setup

pip install -r requirements.txt

1. Calculate sensitive channels

To get sensitive channels, run:

python SNNL.py

remember to input your own pre-trained CNN model at line #379.

2. Prune and fine-tune the model

Using train_cnn.py to train and fine-tune your own model to achieve better fairness.

python train_cnn.py

remember to input your pre-trained CNN model at line #208 and the indexes of sensitive channels at line #212, which can be obtained from the last step where sensitive channels were calculated.

3. Retrain

You can repeat step 1-2 for several times (until the stopping criteria are met) to achieve better results.

Citation

If it is helpful to you, please cite our work:

@inproceedings{kong2024achieving,
  title={Achieving Fairness Through Channel Pruning for Dermatological Disease Diagnosis},
  author={Kong, Qingpeng and Chiu, Ching-Hao and Zeng, Dewen and Chen, Yu-Jen and Ho, Tsung-Yi and Hu, Jingtong and Shi, Yiyu},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={24--34},
  year={2024},
  organization={Springer}
}