PFA_pytorch

This is the source code of paper "Projected Federated Averaging with Heterogeneous Differential Privacy" (accepted by VLDB 2022).

But the original is not this, if you need to see the original, please see https://github.com/Emory-AIMS/PFA

Getting Started

This repository is an implementation of Pytorch version of Federated Averaging (FedAvg), Projected Federated Averaging (PFA) and Projetced Federated Averaging Plus (PFA plus) algorithms.

Install

Pytroch = 1.12.1
Opacus = 1.4.0

Usage

• NP-FedAvg algorithm:

python main.py --Fedavg=True

• DP-FedAvg algorithm:

python main.py --Fedavg=True --dp=True

• PFA algorithm:

python main.py --PFA=True --dp=True

• PFA plus algorithm:

python main.py --PFA_plus=True --dp=True

Acknowledgements

• Pytorch
• Pytorch/Opacus

Contributing

The first author for this job is Junxu Liu. If you have any questions about this article, please contact junxu_liu@ruc.edu.cn

If you have any questions about this code, please email me zl16035056@163.com