ArtFL Offical Source Code

ArtFL: Exploiting Data Resolution in Federated Learning for Dynamic Runtime Inference via Multi-Scale Training

News

[Paper 🤗] [Demo Video🤗]

• 🔥 Upload the training record of CIFAR-10
• 🏆 ArtFL got Best Paper Award at IPSN 2024
• 🔥 Release Sample Code

Install

1. Clone the repo into a local folder. ` git clone https://github.com/siyang-jiang/ArtFL.git

cd ArtFL `

2. Install packages.

conda env create -f environment.yml
conda activate exFL

Quick Start

Before run the bash, make sure you have already prepare the right path to the dataset.

• Change the data path: utils/utils.py change the path and make the data right

bash run.sh

See the training log in CIFAR-10

tensorboard --log_dir=./exp_results_cifar10

Source Tree

├── config
├── data
├── dataloader
│   ├── __init__.py
│   ├── LoaderCifar_deprecated.py
│   ├── LoaderCifar.py
│   └── loader_mini_imagenet.py
├── dataset
│   ├── cifar10.py
│   ├── cifar.py
│   └── tiny_imagenet.py
├── environment.yml
├── fed_branch.py
├── LICENSE
├── loss
├── models
├── README.md
├── requirements.txt
├── run.sh
├── src
├── train_branch.py
└── utils

Citation

Acknowledgement

BalanceFL (IPSN 2022)

License

MIT License.