Haoxiang-Wang / FedADG

Federated Adversarial Domain Generalization

Fedrated Learning with Domain Generalization

Code to reproduce the experiments of Fedrated Learning with Domain Generalization.

How to use it

• Clone or download the repository

Install the requirement

pip install -r requirements.txt

Download VLCS and PACS

• Download VLCS from https://github.com/jindongwang/transferlearning/blob/master/data/dataset.md, extract, move it to ./data/VLCS/
• Download PACS from https://drive.google.com/uc?id=0B6x7gtvErXgfbF9CSk53UkRxVzg, move it to ./data/PACS/
• Download pre-trained AlexNet from https://drive.google.com/file/d/1wUJTH1Joq2KAgrUDeKJghP1Wf7Q9w4z-/view?usp=sharing and move it to ./data/

Running ours on VLCS

cd src
python train.py --lr0 0.001 --lr1 0.0007 --label_smoothing 0.2 --lr-threshold 0.0001 --factor 0.2 --epochs 10 --rp-size 1024 --patience 20 --ite-warmup 500 --global_epochs  20

Running baseline (FedAVG) on VLCS

cd src
python trainFedLear.py --lr0 0.001 --label_smoothing 0.2 --lr-threshold 0.0001 --factor 0.2 --epochs 10 --patience 20 --ite-warmup 500 --global_epochs  20

Running ours on PACS

cd pacs-ours
python train.py --lr0 0.01 --lr1 0.007 --label_smoothing 0.01 --lr-threshold 0.0001 --factor 0.2 --epochs 7 --i-epochs 3 --rp-size 1024 --patience 20 --ite-warmup 100 --global_epochs  30

Running baseline (FedAVG) on PACS

cd pacs-ours
python trainFedLear.py --lr0 0.0001 --label_smoothing 0.01 --lr-threshold 0.0001 --factor 0.2 --epochs 10 --patience 20 --ite-warmup 100 --global_epochs  30