Fast-FedUL

Repo for 'Fast-FedUL: A Training-Free Federated Unlearning with Provable Skew Resilience'

Other baselines can be found at: https://github.com/tamlhp/awesome-machine-unlearning

Citation

Please read and cite our paper:

Huynh, T.T., Nguyen, T.B., Nguyen, P.L., Nguyen, T.T., Weidlich, M., Nguyen, Q.V.H. and Aberer, K., 2024. Fast-FedUL: A Training-Free Federated Unlearning with Provable Skew Resilience. arXiv preprint arXiv:2405.18040.

@article{huynh2024fast,
  title={Fast-FedUL: A Training-Free Federated Unlearning with Provable Skew Resilience},
  author={Huynh, Thanh Trung and Nguyen, Trong Bang and Nguyen, Phi Le and Nguyen, Thanh Tam and Weidlich, Matthias and Nguyen, Quoc Viet Hung and Aberer, Karl},
  journal={arXiv preprint arXiv:2405.18040},
  year={2024}
}

QuickStart

First, run the command below to get the ARDIS dataset (which is used as backdoor data for MNIST dataset):

# change to the ARDIS_DATASET_IV folder
cd customdata/ARDIS_DATASET_IV
# unrar ARDIS_DATASET
unrar e ARDIS_DATASET_IV.rar
# return to the root folder
cd ...

Second, run the command below to get the splited dataset MNIST:

bash gen_data.sh

The splited data will be stored in ./fedtask/mnist_cnum25_dist0_skew0_seed0/attack.json.

Third, run the command below to quickly run the experiment on MNIST dataset:

# all parameters are set in the file run_exp.sh
bash run_exp.sh

The result will be stored in ./fedtasksave/mnist_cnum25_dist0_skew0_seed0/R51_P0.30_alpha0.07/record/history51.pkl.

Finally, run the command below to return accuracy:

# return accuracy
python test_unlearn.py
# Main accuracy: ...
# Backdoor accuracy: ...