FLATS: Federated Learning Adversarial Training for Smart Home Face Recognition System

• This is the official repository of the code of "Robust Smart Home Face Recognition under Starving Federated Data" paper, which has been accepted by the IEEE International Conference on Universal Village (IEEE UV2022).
• We propose a novel robust federated learninng training method for smart home face recognition system named FLATS: Federated Learning Adversarial Training for Smart Home Face Recognition System.
• For general overview of the training process, take a look at the notebook/FLATS.ipynb file.
• The code will be continously updated.
• Oral Presentation Link: https://youtu.be/Tj9QiJEUBXU

LICENSES

• Includes software related under the MIT and Apache 2.0 license

Run

Arguments to be parsed:

• --main_folder_path
• --num_clients (default=5)
• --train_batch_size (default=64)
• --test_batch_size (default=64)
• --num_selected (default=5)
• --num_attack (default=1)
• --num_rounds (default=10)
• --num_local_epochs (default=5)
• --clean_train_batch_ratio (default=5)
• --atk (default=FFGSM(white_model, eps=8/255, alpha=10/255))

Run:

CUDA_VISIBLE_DEVICES=0 python main.py \
  --main_folder_path 'pins_face_recognition_105_classes' \ 
  --num_clients 5 \ 
  --train_batch_size 64 \ 
  --test_batch_size 64 \ 
  --num_selected 5 \ 
  --num_attack 1 \ 
  --num_rounds 10 \ 
  --num_local_epochs 5 \ 
  --clean_train_batch_ratio 5 \ 
  --atk FFGSM(white_model, eps=8/255, alpha=10/255)