Official repository for WWW 2022 paper An Accuracy-Lossless Perturbation Method for Defending Privacy Attacks in Federated Learning. This project is developed based on Python 3.6.

Install prerequisites

pip install -r requirements.txt

Download datasets

Download the "" [download link] and run 'unzip dataset.zip' at the root directory before training.

Regression experiments

All code for this part are included in "UMBD" and "ABD" subfolder. Please change to that folder before running the code.

Running demos

PBPFL training

• Train models with PBPFL

bash pbpfl_train.sh

AVGFL training

• Train models with traditional AVGFL

bash avgfl_train.sh

Classification experiments

All code for this part are included in "LDC" subfolder. Please change to that folder before running the code.

Running demos

PBPFL training

• Train models with PBPFL

bash pbpfl_train.sh

AVGFL-MSE training

• Train models with traditional AVGFL on MSE loss

bash avgfl_mse_train.sh

AVGFL training

• Train models with traditional AVGFL on CE loss

bash avgfl_ce_train.sh

Citation

Please cite our paper in your publications if it helps your research:

@inproceedings{Yang_WWW_2022,
  title={An Accuracy-Lossless Perturbation Method for Defending Privacy Attacks in Federated Learning},
  author={Xue Yang, Yan Feng, Weijun Fang, Jun Shao, Xiaohu Tang, Shu-Tao Xia, Rongxing Lu},
  booktitle={Proceedings of the ACM Web Conference 2022},
  year={2022}
}