GAS: Generative Activation-Aided Asynchronous Split Federated Learning

This repository is the demo of GAS.

Requirements

To install the required packages:

pip install -r requirements.txt

Training and Evaluation

To train and evaluate the model(s) in the paper, run this command:

python GAS_main.py 

Results

Our model achieves the following performance on CIFAR-10, CIFAR100, CINIC10 and Fashion-MNIST:

Dataset	$s=2$	$\alpha=0.1$
CIFAR10	$82.78±0.58$	$81.72±0.50$
CINIC10	$68.32±0.17$	$65.94±1.14$
Fashion-MNIST	$90.66±0.20$	$90.58±0.34$