Skip to content
/ K-Vib Public

Code for paper - Exploring Federated Optimization by Reducing Variance of Adaptive Unbiased Client Sampling

License

Notifications You must be signed in to change notification settings

dunzeng/K-Vib

Repository files navigation

K-Vib

Code for paper - Exploring Federated Optimization by Reducing Variance of Adaptive Unbiased Client Sampling

Dataset

Our experiment setting of FEMNIST follows Optimal Client Sampling, please download the modified dataset following their instructions. You can download from google driver as well.

They are expected to be located in datasets directory.

Run

synthetic task

python main_synthetic.py -com_round 500 -sample_ratio 0.1 -num_clients 100 -batch_size 64 -epochs 1 -lr 0.02 -dseed 4399 -seed [run_seed] -data synthetic -sampler [uniform/kvib] -reg 0.1

femnist task

python main_femnist.py -com_round 500 -k [num_of_client_per_round] -dataset v2 -batch_size 20 -epochs 3 -lr [learning_rate] -freq 10 -sampler [uniform/kvib] -reg 0.1 -seed [run_seed]

Citation

@article{zeng2023exploring,
  title={Exploring Federated Optimization by Reducing Variance of Adaptive Unbiased Client Sampling},
  author={Zeng, Dun and Xu, Zenglin and Pan, Yu and Wang, Qifan and Tang, Xiaoying},
  journal={arXiv preprint arXiv:2310.02698},
  year={2023}
}

About

Code for paper - Exploring Federated Optimization by Reducing Variance of Adaptive Unbiased Client Sampling

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages