DualOptim

Introduction

Official codes for our paper submitted to IEEE TIFS: Refining Distributed Noisy Clients: An End-to-end Dual Optimization Framework. This paper is now preprinted on Techrxiv via this link.

Installation

Please check below requirements and install packages from requirements.txt.

$ pip install --upgrade pip
$ pip install -r requirements.txt

Usage

The following commands are examples representing symmetric 0.0-0.4, pairflip 0.0-0.4 and mixed 0.0-0.4 label noise settings.

# symmetric 0.0-0.4 dirichlet 1.0 

python main_fed_LNL.py \
--dataset cifar10 \
--model resnet18 \
--epochs 120 \
--noise_type_lst symmetric \
--noise_group_num 100  \
--group_noise_rate 0.0 0.4 \
--partition dirichlet \
--dd_alpha 0.5 \
--method dualoptim | tee  symmetric04_dir05.txt

# pairflip 0.0-0.4 dirichlet 1.0 

python main_fed_LNL.py \
--dataset cifar10 \
--model resnet18 \
--epochs 120 \
--noise_type_lst pairflip \
--noise_group_num 100  \
--group_noise_rate 0.0 0.4 \
--partition dirichlet \
--dd_alpha 0.5 \
--method dualoptim | tee  pairflip04_dir05.txt

# mixed 0.0-0.4
python main_fed_LNL.py \
--dataset cifar10 \
--model resnet18 \
--epochs 120 \
--noise_type_lst symmetric pairflip\
--noise_group_num 50 50  \
--group_noise_rate 0.0 0.4 0.0 0.4 \
--partition dirichlet \
--dd_alpha 1.0 \
--method dualoptim | tee mixed04_cifar10_dir1.txt

Parameters for noisy label

Parameter	Description
noise_type_lst	Noisy type list.
noisy_group_num	Number of clients corresponding to noisy type.
group_noise_rate	The noise rate corresponding to the noisy group. It increases linearly from 0.0 to noise rate for each group.

Please check run.sh for commands for various data and noisy label scenarios.

Citing this work

If you find this work helpful, please consider crediting our work:

 @article{DualOptim,
title={Refining Distributed Noisy Clients: An End-to-end Dual Optimization Framework},
url={http://dx.doi.org/10.36227/techrxiv.173707406.66001019/v1},
DOI={10.36227/techrxiv.173707406.66001019/v1},
publisher={Institute of Electrical and Electronics Engineers (IEEE)},
author={Jiang, Xuefeng and Li, Peng and Sun, Sheng and Li, Jia and Wu, Lvhua and Wang, Yuwei and Lu, Xiuhua and Ma, Xu and Liu, Min},
year={2025},
month=jan
}