|
|
|
- Working through an assignment from CS236 class at Stanford on deep generative models, [homework 3](uploads/1189a481a0cdf4e5eae5e1258fb5b6a5/CS236_Homework_3.pdf), problem 4 – Wasserstein GAN; [starter pack for homework](uploads/8bf57089d4342d57298d10f8559234de/hw3starter.zip)
|
|
|
|
- We were wondering why the regularizer was forcing the gradient to be one. It seems that the paper that introduced WGAN with gradient penalty proved that the gradient of the optimal discriminator (critic) is, in fact, one almost everywhere. |
|
|
\ No newline at end of file |
