Learn How to Query from Unlabeled Data Streams in Federated Learning

This is the implementation of our paper "Learn How to Query from Unlabeled Data Streams in Federated Learning" in AAAI 2025.

Installation instructions

Install Python environment with conda:

conda create -n leadq
conda activate leadq

Then install the Python packages in requirements.txt:

pip install -r requirements.txt

NOTE: you may need to check the version of some packages such as torch.

Run an experiment

python main.py --al_method=[strategy name] --model=[model name] --dataset=[dataset name] --gpu=[gpu id] \
    --n_arrive=[number of arrived unlabeled samples] --n_query=[number of queried samples] 

Argument	Description	Choices
al_method	The data querying strategy	leadq, kafal, logo, coreset
model	The name of the model	cnnvconv, resnet18
dataset	The name of the dataset	svhn, cifar100
gpu	The id of GPU	0,1, ...
n_arrive	The number of arrived samples per client per round	10, 20, ... (Integer)
n_query	The number of queried samples per client per round	1, 2, ... (Integer)

For example, run the proposed method:

python main.py --al_method=leadq --model=cnn4conv --dataset=svhn --gpu=0 --n_arrive=10 --n_query=1

Citing

If you use this code in your research or find it helpful, please consider citing our paper:

@article{sun2025learn,
  title={Learn How to Query from Unlabeled Data Streams in Federated Learning},
  author={Sun, Yuchang and Li, Xinran and Lin, Tao and Zhang, Jun},
  booktitle={accepted by The 39th Annual AAAI Conference on Artificial Intelligence (AAAI)},
  year={2025}
}

Contact

If you have any questions, please feel free to contact us via [email protected].

Acknowledgements

The initial implement of this repo is based on LoGo. We thank the authors for their contribution.