This repository contains the code to reproduce the Follow-the-Ridge results from the paper On Solving Minimax Optimization Locally: A Follow-the-Ridge Approach.
Particularly, you can reproduce our results of GAN on 1-D and 2-D mixture of Gaussians.
# for 1-D MOG
$ python follow_ridge_1D.py --follow_ridge --adapt_damping
# for 2-D MOG
$ python follow_ridge_2D.py --follow_ridge --adapt_damping
You can also experiment with three low dimensional toy examples.
# for function g1
$ python quadratic.py --function 1
Python 3.6, Tensorflow 1.14.0
To cite this work, please use
@inproceedings{
wang2020on,
title={On Solving Minimax Optimization Locally: A Follow-the-Ridge Approach},
author={Yuanhao Wang and Guodong Zhang and Jimmy Ba},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=Hkx7_1rKwS}
}