FedADMM: A Robust Federated Deep Learning Framework with Adaptivity to System Heterogeneity
Author: Yonghai Gong, Yichuan Li, and Nikolaos M. Freris
The code is modified from https://github.com/AshwinRJ/Federated-Learning-PyTorch
Install all the packages from requirments.txt
- python = 3.8.5
- pytorch = 1.9.0
- torchvision = 0.10.0
- numpy = 1.19.2
- tensorboardx = 2.2
- matplotlib = 3.3.2
| Model | # of para | Dataset | Target Accuracy |
|---|---|---|---|
| CNN 1 | 1,663,370 | MNIST | 97% |
| CNN 1 | 1,663,370 | FMNIST | 80% |
| CNN 2 | 1,105,098 | CIFAR-10 | 45% |
IID:data are evenly distributed to clientsnon-IID:first arrange the training data by label and then distribute them into shards. Each client is assigned two shards randomly.
The default values for various parameters parsed to the experiment are given in options.py. Details are given on some of those parameters:
--num_users: number of users: m--frac:the fraction of clients: C--epochs:number of global rounds of training--local_ep:the number of local epochs: E--local_bs:local batch size: B--eta:server step size--lr:local learning rate--rho:coefficient of the proximal parameter:--iid:1 for IID, and 0 for non-IID--model:name of the model--dataset:name of the dataset
If you find this repository useful for your research, please cite the following paper:
@inproceedings{fedadmm,
title={Fed{ADMM}: A Robust Federated Deep Learning Framework with Adaptivity to System Heterogeneity},
author = {Gong, Yonghai and Li, Yichuan and Freris, Nikolaos M.},
booktitle = {ICDE},
pages = {2575-2587},
year = {2022}
}
(If you have any questions, feel free to ask me via gongyh@mail.ustc.edu.cn)