FedTune

Source code for FedTune.

The core code for FedTune is in fedtuner.py

Dataset Download and Preprocess

Please refer to Dataset/README.md

Experiments

1. FL training with FedTune enabled

   CUDA_VISIBLE_DEVICES=7 python Project/FedTune/main.py --enable_fedtune False --aggregator fednova --model resnet_10 --target_model_accuracy 0.8 --n_participant 20 --n_training_pass 20 --dataset speech_commands

2. FL training with FedTune disabled

   CUDA_VISIBLE_DEVICES=7 python Project/Fedtune/main.py --enable_fedtune False --aggregator fedavg --model resnet_10 --target_model_accuracy 0.8 --n_participant 20 --n_training_pass 20 --dataset speech_commands 

   Required arguments:

   • --enable_fedtune False
   • --aggregator
   • --model
   • --target_model_accuracy
   • --dataset
   • --n_participant
   • --n_training_pass

3. Optional arguments

   • --n_consecutive_better: number of trained model is consecutively better than the target accuracy before stop training. Default 5.
   • --trace_id: trace id. Default 1.
   • --penalty: penalty factor when bad decision occurs. Default 10.