Implementation of the paper No Fear of Heterogeneity: Classifier Calibration for Federated Learning with Non-IID Data

Run this repo:

1. Download the cifar10 dataset and save as images in the dir "./data/"

   python data_process.py

2. Run the main procedure:

   python main.py

3. Run t-SNE visualization:

   python visualize.py [--model_before_calibration MODEL_BEFORE_CALIBRATION] [--model_after_calibration MODEL_AFTER_CALIBRATION] [--random_state RANDOM_STATE] [--save_path SAVE_PATH]

   Default arguments are:

   • MODEL_BEFORE_CALIBRATION: ./save_model/model-epoch9.pth
   • MODEL_AFTER_CALIBRATION: ./save_model/model.pth
   • RANDOM_STATE: 1
   • SAVE_PATH: ./visualize/tsne.png