Embracing Federated Learning: Enabling Weak Client Participation via Partial Model Training

Sunwoo Lee, Tuo Zhang, Saurav Prakash, Yue Niu, Salman Avestimehr

IEEE Transactions on Mobile Computing (TMC), 2024

This work proposes a federated learning framework that lets memory-constrained clients participate by having each client train only as many consecutive output-side layers as its resources allow, provably converging near stationary points for non-convex problems and matching strong-client accuracy across CIFAR-10, FEMNIST, and IMDB while outperforming state-of-the-art width-reduction methods like HeteroFL and FjORD.

BibTeX

@article{lee2024embracing,
  title={Embracing federated learning: Enabling weak client participation via partial model training},
  author={Lee, Sunwoo and Zhang, Tuo and Prakash, Saurav and Niu, Yue and Avestimehr, Salman},
  journal={IEEE Transactions on Mobile Computing},
  volume={23},
  number={12},
  pages={11133--11143},
  year={2024},
  publisher={IEEE}
}