This is the Pytorch implementation of our paper, PrivateFL: Accurate, Differentially Private Federated Learning via Personalized Data Transformation.
First enter the following path:
cd scriptWe use miniconda to create a virtual environment with python 3.8, you can install miniconda use the following script if you are using Linux-x86-64bit machine:
(Optional for install miniconda)
bash install_conda.shThen use the following script to download the requirements:
bash setup.shYou can use the following script to train from scratch.
bash fedavg.shYou can also change the parameters in script/train.sh, e.g., --data --nclient --nclass --ncpc --model --mode --round --epsilon --sr --lr, following the choices listed in parse_arguments() of FedAverage.py. The value of the parameters can be found in our paper.
Run the following script to extract features from [ResNeXt, SimCLR, CLIP] and train a one-layer classifier, you may need to download ResNext using this link:
bash fedtransfer.shPlease reduce the value of --physical_bs if facing CUDA out of memory.
@inproceedings{yangprivatefl,
title={PrivateFL: Accurate, Differentially Private Federated Learning via Personalized Data Transformation},
author={Yang, Yuchen and Hui, Bo and Yuan, Haolin and Gong, Neil and Cao, Yinzhi}
booktitle = {Proceedings of the USENIX Security Symposium (Usenix'23)},
year = {2023}
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent