benrogboe / symplectic-gradient-adjustment

A colab that implements the Symplectic Gradient Adjustment optimizer from "The mechanics of n-player differentiable games"

This colab implements the Symplectic Gradient Adjustment (SGA) algorithm. It then applies this optimizer to two different experiments:

• using a GAN to learn a 4 by 4 mixture of Gaussians;
• using a GAN to learn 75 dimensional Gaussian.

All experiments run in 5-10 minutes with a GPU, but can of course be much slower on CPU.

Use this link to run this notebook in colab: https://colab.research.google.com/github/deepmind/symplectic-gradient-adjustment/blob/master/Symplectic_Gradient_Adjustment.ipynb