FedNN

Pattern Recognition. "FedNN: Federated learning on concept drift data using weight and adaptive group normalizations"

Train

# FedNN
python main.py \
    --method fedavg \
    --data_dir ./data \
    --dataset_name DIGIT \
    --model_name LeNet_fednn \
    --result_path results

# naive FedAvg
python main.py \
    --method fedavg \
    --data_dir ./data \
    --dataset_name DIGIT \
    --model_name LeNet \
    --result_path results

Data

Please download the pre-processed digit datasets here FedBN and put it in the ./data directory.

Requirements

pip install torch torchvision

Citation

If you find this repository useful in your research, please cite:

@article{kang2023fednn,
  title={FedNN: Federated learning on concept drift data using weight and adaptive group normalizations},
  author={Kang, Myeongkyun and Kim, Soopil and Jin, Kyong Hwan and Adeli, Ehsan and Pohl, Kilian M and Park, Sang Hyun},
  journal={Pattern Recognition},
  volume={149},
  pages={110230},
  year={2023},
  publisher={Elsevier}
}

Thanks to works below for their implementations which were useful for this work. FedDC