paulzhengfr / fedproxper

Official code of the paper "Federated Learning in Heterogeneous Networks with Unreliable Communication"

Fedprox PER (packet error rate)

This code consists in federated learning under transmission packet error rate for the paper "Federated Learning in Heterogeneous Networks with Unreliable Communication".

The main program is main_fed.py.

Please install pytorch and related first and then install the requirement.txt file.

You can run the program by:

python main_fed.py --dataset mnist --model Mnist_oldMLP --round 200 --gpu 0 --iid niid --optimizer fedprox --total_UE 500 --active_UE 10 --selection "weighted_random"  --name 'mnist_WR' --local_ep 20 --lr 0.1 --scenario woPER

Many options of the training can be found at the argparse part in the main program.

• --optimizer: FedAvg and FedProx are implemented and can be changed: fedavg, fedprox.
• --iid: type of non-iid: IID for iid, NIID for 2 digits per client, and Dirichlet for dirichlet non-iid distribution tuned by --alpha (represent the similarity of data).
• --batch_size: batch size.
• --lr: learning rate.
• --mu: FedProx parameter.
• --round: total communication rounds to run
• --epochs: local computation
• --seed and --wireless_seed:random seeds
• --total_UE: total participating users
• --active_UE: number of active users at each round.
• --selection: user selection strategy weighted_random, and our method is denoted as solve_opti_loss_size2.
• --name: experiment name for folder name where all results are stored.

More options can be found in ./utils/options.py.

All simulations in the paper can be reproced by their corresponding bash scripts in the folder bash_scripts/. All figures can be reproduced by Fig_X_*.ipynb/py file in this main folder.

Results csv files to reproduce the figures will all be updated after paper official publication.

The paper reference is as belows:

@ARTICLE{10253642,
  author={Zheng, Paul and Zhu, Yao and Hu, Yulin and Zhang, Zhengming and Schmeink, Anke},
  journal={IEEE Transactions on Wireless Communications}, 
  title={Federated Learning in Heterogeneous Networks With Unreliable Communication}, 
  year={2024},
  volume={23},
  number={4},
  pages={3823-3838},
  keywords={Training;Convergence;Computational modeling;Servers;Resource management;Data models;Wireless networks;Distributed learning;federated learning;wireless networks;packet error rate;convergence analysis},
  doi={10.1109/TWC.2023.3311824}}