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.

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.

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.

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}
}