← Publications

Partial Model Averaging in Federated Learning: Performance Guarantees and Benefits

Sunwoo Lee, Anit Sahu, Chaoyang He, Salman Avestimehr

Neurocomputing, 2023

• DOI

This work proposes a partial model averaging framework for Federated Learning that reduces the model discrepancy caused by periodic full averaging in FedAvg-style local SGD, achieving up to 2.2% higher accuracy than periodic full averaging on CIFAR-10/100 and FEMNIST with 128 clients under a fixed training budget.

BibTeX

@article{lee2023partial,
  title={Partial model averaging in federated learning: Performance guarantees and benefits},
  author={Lee, Sunwoo and Sahu, Anit Kumar and He, Chaoyang and Avestimehr, Salman},
  journal={Neurocomputing},
  volume={556},
  pages={126647},
  year={2023},
  publisher={Elsevier}
}