Skip to content

CityU-AIM-Group/PRR-Imbalance

Repository files navigation

Personalized Retrogress-Resilient FL for Imbalanced Medical Data (PRR-Imbalance)

This repository is an official PyTorch implementation of the paper "Personalized Retrogress-Resilient Federated Learning Towards Imbalanced Medical Data" [paper] from IEEE Transactions on Medical Imaging (TMI) 2022.

Download

The dermoscopic FL dataset can be downloaded from Google Drive. Put the downloaded clientA, clientB, clientC and clientD subfolders in a newly-built folder ./data/.

Dependencies

  • Python 3.7
  • PyTorch >= 1.7.0
  • numpy 1.19.4
  • scikit-learn 0.24.2
  • scipy 1.6.2
  • albumentations 0.5.2

Code

Clone this repository into any place you want.

git clone https://github.com/CityU-AIM-Group/PRR-Imbalance.git
cd PRR-Imbalance
mkdir data

Quickstart

  • Train the PRR-Imbalance with default settings:
python ./main.py --theme prr-imbalance --iters 50 --wk_iters 5 --network vgg_nb --l_rate 0.7 --lr 1e-2 

Cite

If you find our work useful in your research or publication, please cite our work:

@ARTICLE{2022personalizedFL,
  title={Personalized Retrogress-Resilient Federated Learning Towards Imbalanced Medical Data}, 
  author={Chen, Zhen and Yang, Chen and Zhu, Meilu and Peng, Zhe and Yuan, Yixuan},
  journal={IEEE Transactions on Medical Imaging}, 
  year={2022},
  pages={1-1},
  doi={10.1109/TMI.2022.3192483}
}

About

[TMI'22] Personalized Retrogress-Resilient Federated Learning Towards Imbalanced Medical Data

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published