FedPU — Official PyTorch Implementation

For any inquiries, please contact Xinyang Lin at [email protected]

ICML2022 - Federated Learning with Positive and Unlabeled Data

Using the Code

Requirements

This code has been developed under Python3.7, PyTorch 1.7.0 and CUDA 11.0 on Red Hat 8.3.

Training

Train FedPU model. We implemented the dataloader of FedMatch(ICLR 2021) on cifar10, for easier comparison.

FedPU works with FedAvg on non-iid data:

sh train_c10_FedAvg_FedPU_fmloader_noniid.sh

FedPU works with FedProx on non-iid data:

sh train_c10_FedProx_FedPU_fmloader_noniid.sh

Supervised learning experiment can be performed:

sh train_c10_FedAvg_SL_fmloader_noniid.sh

Citation

If our work is helpful for your research, please consider citing:

@inproceedings{lin2022federated,
  title={Federated Learning with Positive and Unlabeled Data},
  author={Lin, Xinyang and Chen, Hanting and Xu, Yixing and Xu, Chao and Gui, Xiaolin and Deng, Yiping and Wang, Yunhe},
  booktitle={International Conference on Machine Learning},
  pages={13344--13355},
  year={2022},
  organization={PMLR}
}