AugHFL

This repository provides resources for the following paper:

[Robust Heterogeneous Federated Learning under Data Corruption](ICCV 2023 Open Access Repository)
Xiuwen Fang, Mang Ye, Xiyuan Yang ICCV 2023

AugHFL Description

AugHFL(Augmented Heterogeneous Federated Learning) is a federated learning framework to investigate the problem of data corruption in the model heterogeneous federated learning:

1. Local Learning with Data Corruption.

2. Robust Corrupted Clients Communication.

Framework Architecture

Dataset

Our experiments are conducted on two datasets, Cifar-10-C and Cifar-100. We set public dataset on the server as a subset of Cifar-100, and randomly divide Cifar-10-C to different clients as private datasets.

Dataset used: CIFAR-10-C、CIFAR-100

Note: Cifar-10-C can be genearetd by make_cifar_c.py. Data will be processed in init_data.py.

Quick Start

You can start training and evaluation as follows:

pip install -r requirements.txt
# init public data and local data
python Dataset/init_data.py
# pretrain local models
python Network/pretrain.py
# AugHFL
python AugHFL/AugHFL.py

Script and Sample Code

├── Robust_FL
    ├── Dataset
        ├── augmentations.py
        ├── cifar.py
        ├── dataaug.py
        ├── init_dataset.py
        ├── utils.py
    ├── Network
        ├── Models_Def
            ├── mobilnet_v2.py
            ├── resnet.py
            ├── shufflenet.py
        ├── pretrain.py
    ├── AugHFL
        ├── AugHFL.py
    ├── loss.py
    ├── README.md

Comparison with the SOTA methods

In the heterogeneous model scenario, we assign four different networks:ResNet10,ResNet12,ShuffleNet,Mobilenetv2

Citation

@inproceedings{fang2023robust,
  title={Robust heterogeneous federated learning under data corruption},
  author={Fang, Xiuwen and Ye, Mang and Yang, Xiyuan},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={5020--5030},
  year={2023}
}