A PyTorch implementation for the paper FedCon: A Contrastive Framework for Federated Semi-Supervised Learning.

We have 5 baselines (FedAvg-FixMatch, FedProx-FixMatch, FedAvg-UDA, FedProx-UDA, FedMatch, SSFL) and our proposed FedCon framework in our experiment.

We do our experiments on MNIST, CIFAR-10, and SVHN datasets.

you should place your data in ./fedcon-ecmlpkdd2021/data/mnist (mnist for example)

python>=3.6 pytorch>=0.4

To install PyTorch, see installation instructions on the PyTorch website.

We provide some examples here.

python fedcon-main.py --data_dir ../data/mnist --backbone Mnist --dataset mnist --batch_size 10 --num_epochs 200 --label_rate 0.01 --iid iid

python fedcon-main.py --data_dir ../data/mnist --backbone Mnist --dataset mnist --batch_size 10 --num_epochs 200 --label_rate 0.1 --iid iid

python fedcon-main.py --data_dir ../data/mnist --backbone Mnist --dataset mnist --batch_size 10 --num_epochs 200 --label_rate 0.01 --iid noniid

python fedcon-main.py --data_dir ../data/cifar --backbone Cifar --dataset cifar10 --batch_size 10 --num_epochs 200 --label_rate 0.01 --iid iid

python fedcon-main.py --data_dir ../data/svhn --backbone Svhn --dataset svhn --batch_size 10 --num_epochs 150 --label_rate 0.01 --iid iid

You can cite our work by

@article{long2021fedcon,
  title={FedCon: A Contrastive Framework for Federated Semi-Supervised Learning},
  author={Long, Zewei and Wang, Jiaqi and Wang, Yaqing and Xiao, Houping and Ma, Fenglong},
  journal={arXiv preprint arXiv:2109.04533},
  year={2021}
}