eceisik / ARFED

Byzantine Attack-Resistant Federated Averaging Based on Outlier Elimination

ARFED: Attack Resistant Federated Averaging Based on Outlier Elimination

This repository is the official implementation of ARFED: Attack Resistant Federated Averaging Based on Outlier Elimination. The framework and full methodology are detailed in our manuscript.

Install required packages

pip install -r requirements.txt

for pip environment. OR

conda env create -f environment.yml

To create a conda virtual environment from an environment.yml file. Please see here for more details.

How to run

We performed the experiments for label flipping and Byzantine attacks on MNIST, Fashion MNIST, and CIFAR10 datasets. We compared our results with the median method. The file name list for each data set is as follows. The default setting of experiments is NON-IID organized with a 20% malicious participant ratio.

Dataset	Attack Type	Method	File Name
MNIST	Byzantine	ARFED	mnist_byz_barfed.py
MNIST	Byzantine	Median	mnist_byz_median.py
MNIST	Byzantine	Trimmed Mean	mnist_byz_trimmed.py
MNIST	Byzantine	No Defense	mnist_byz_attack.py
MNIST	Label Flipping	ARFED	mnist_lf_barfed.py
MNIST	Label Flipping	Median	mnist_lf_median.py
MNIST	Label Flipping	Trimmed Mean	mnist_lf_trimmed.py
MNIST	Label Flipping	No Defense	mnist_byz_attack.py
MNIST	Partial Knowledge	ARFED	mnist_partial_knowledge_barfed.py
MNIST	Partial Knowledge	Median	mnist_partial_knowledge_median.py
MNIST	Partial Knowledge	Trimmed Mean	mnist_partial_knowledge_trimmed.py
MNIST	Partial Knowledge	No Defense	mnist_partial_knowledge_attack.py
Fashion MNIST	Byzantine	ARFED	fashion_byz_barfed.py
Fashion MNIST	Byzantine	Median	fashion_byz_median.py
Fashion MNIST	Byzantine	Trimmed Mean	fashion_byz_trimmed.py
Fashion MNIST	Byzantine	No Defense	fashion_byz_attack.py
Fashion MNIST	Label Flipping	ARFED	fashion_lf_barfed.py
Fashion MNIST	Label Flipping	Median	fashion_lf_median.py
Fashion MNIST	Label Flipping	Trimmed Mean	fashion_lf_trimmed.py
Fashion MNIST	Label Flipping	No Defense	fashion_lf_attack.py
Fashion MNIST	Partial Knowledge	ARFED	fashion_partial_knowledge_barfed.py
Fashion MNIST	Partial Knowledge	Median	fashion_partial_knowledge_median.py
Fashion MNIST	Partial Knowledge	Trimmed Mean	fashion_partial_knowledge_trimmed.py
Fashion MNIST	Partial Knowledge	No Defense	fashion_partial_knowledge_attack.py
CIFAR10	Byzantine	ARFED	cifar_byz_barfed.py
CIFAR10	Byzantine	Median	cifar_byz_median.py
CIFAR10	Byzantine	Trimmed Mean	cifar_byz_trimmed.py
CIFAR10	Byzantine	No Defense	cifar_byz_attack.py
CIFAR10	Label Flipping	ARFED	cifar_lf_barfed.py
CIFAR10	Label Flipping	Median	cifar_lf_median.py
CIFAR10	Label Flipping	Trimmed Mean	cifar_lf_trimmed.py
CIFAR10	Label Flipping	No Defense	cifar_lf_attack.py
CIFAR10	Partial Knowledge	ARFED	cifar_partial_knowledge_barfed.py
CIFAR10	Partial Knowledge	Median	cifar_partial_knowledge_median.py
CIFAR10	Partial Knowledge	Trimmed Mean	cifar_partial_knowledge_trimmed.py
CIFAR10	Partial Knowledge	No Defense	cifar_partial_knowledge_attack.py

Citation

Please use the following BibTeX entry for citation:

@article{ISIKPOLAT2023626,
title = {ARFED: Attack-Resistant Federated averaging based on outlier elimination},
journal = {Future Generation Computer Systems},
volume = {141},
pages = {626-650},
year = {2023},
issn = {0167-739X},
doi = {https://doi.org/10.1016/j.future.2022.12.003},
url = {https://www.sciencedirect.com/science/article/pii/S0167739X22004083},
author = {Ece Isik-Polat and Gorkem Polat and Altan Kocyigit}
}