Federated Learning Simulator on Light Tasks

We include several most common used federated tasks in this benchmark. You can validate and compare different federated algorithms under various experiment settings efficiently.

Install

git clone git@github.com:FederalLab/benchmark-lightly.git
cd benchmark-lightly
pip install -r requirements.txt
python setup.py install

Benchmark

Refer to docs/ for more details. We provide a flexible settings configuration for benchmark, rather than given an specified settings, since different settings will have a large impact on performance. It is more convenient to validate different algorithms in there ideal situation, such as non-iid or iid distribution, full-activated or partially activated clients.

Citation

If you find this project useful in your research, please consider cite:

@misc{chen2021openfed,
      title={OpenFed: A Comprehensive and Versatile Open-Source Federated Learning Framework},
      author={Dengsheng Chen and Vince Tan and Zhilin Lu and Jie Hu},
      year={2021},
      eprint={2109.07852},
      archivePrefix={arXiv},
      primaryClass={cs.CR}
}