superjdz / CGL-GAN

A distributed learning algorithm for GANs with Non-IID dataset

A Novel Federated Learning Scheme for Generative Adversarial Networks

Collaborative Game Parallel Learning

A distributed learning algorithm for GANs with Non-IID dataset

Env-requirements

• torch
• matplotlib
• fedlab

Comparison Algorithm

• AC-GAN
• FL-GAN
• MD-GAN
• FeGAN
• CAP-GAN (ours)

Usage

All codes use global variants to control the experiment setting Regular parameters

num_workers :   The number of clients in the system
num_servers :   The number of edge servers in the system
E : To control when to share the discriminator to neighbors. Note that: We provide the code in notes and just cancel the note if you want to test it
num_class :     This parameter only acts on generating how many classes of 2D-Gaussian Mixture 
num_sample : To control the number of samples for testing
batch_size : Batch Size
frac_workers: It is for FL-GAN or FeGAN
epoch:  The epoch for local iterations in clients.

ACGAN

Enter /ACGAN/2DMG (MNIST)/

python acgan.py

MDGAN

Enter /MDGAN/2DMG (MNIST)/

python mdgan.py

FeGAN

python fegan.py

CAPGAN

Enter /CAPGAN/MNIST (MNIST)/

python main.py

Enable Mix-G module (mixgan) Enter the /CAPGAN/MNIST/

python mixed-gan.py