LEAD: Min-Max Optimization from a Physical Perspective

This is the code associated with the paper "LEAD: Min-Max Optimization from a Physical Perspective". If you find this code useful please cite us:

@article{
askari hemmat2023lead,
title={{LEAD}: Min-Max Optimization from a Physical Perspective},
author={Reyhane Askari Hemmat and Amartya Mitra and Guillaume Lajoie and Ioannis Mitliagkas},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2023},
url={https://openreview.net/forum?id=vXSsTYs6ZB},
note={Featured Certification}
}

Video describing the paper: https://www.youtube.com/watch?v=EfwIc0GXb8E

Blogpost: https://reyhaneaskari.github.io/LEAD.html

For any questions about the code please create an issue.

The code requires pytorch and tensorflow. But TF is only used for computing the inception score.

Acknowledgement

1. DCGAN code adpoted from https://github.com/Zeleni9/pytorch-wgan
2. ResNet code adopted from https://github.com/GongXinyuu/sngan.pytorch
3. SGA code implemented based on https://github.com/deepmind/symplectic-gradient-adjustment/blob/master/Symplectic_Gradient_Adjustment.ipynb
4. Extra-Adam optim source code from https://github.com/GauthierGidel/Variational-Inequality-GAN
5. CGD optim source code from https://github.com/devzhk/Implicit-Competitive-Regularization