gong-xuan / fedkd

Federated Learning w/ Knowledge Distillation

This repository contains the implementation for Preserving Privacy in Federated Learning with Ensemble Cross-Domain Knowledge Distillation (AAAI 2022).

Contact: Xuan Gong ([email protected])

Federated distillation with multi communication rounds

python main.py --alpha 1.0 --seed 1 --C 1

Oneshot federated distillation

python main.py --alpha 1.0 --seed 1 --C 1 --oneshot

Oneshot federated distillation with quantization and noise

python main.py --alpha 1.0 --seed 1 --C 1 --oneshot --noisescale 1.0 --quantify 100
python main.py --alpha 0.1 --seed 1 --C 1 --oneshot --noisescale 1.0 --quantify 100
python main.py --alpha 1.0 --seed 1 --C 1 --dataset cifar100 --oneshot --noisescale 1.0 --quantify 100
python main.py --alpha 0.1 --seed 1 --C 1 --dataset cifar100 --oneshot --noisescale 1.0 --quantify 100

Citation

@inproceedings{gong2022preserving,
  title={Preserving Privacy in Federated Learning with Ensemble Cross-Domain Knowledge Distillation},
  author={Gong, Xuan and Sharma, Abhishek and Karanam, Srikrishna and Wu, Ziyan and Chen, Terrence and Doermann, David and Innanje, Arun},
  booktitle = {Proceedings of the AAAI conference on Artificial Intelligence},
  pages     = {11891--11899},
  year      = {2022},
}