Oleg Zabluda's blog
Tuesday, November 28, 2017
 
Read-through: Wasserstein GAN
Read-through: Wasserstein GAN
"""
Aside: What’s Up With The Earth Mover Definition?
[...]
We’d like our optimization algorithm to learn to move θ to 0, As θ→0, the distance d(P​0,Pθ) should decrease. But for many common distance functions, this doesn’t happen.
[...]
If we remove batch norm from the generator, WGAN still generates okay samples, but DCGAN fails completely.
"""
https://www.alexirpan.com/2017/02/22/wasserstein-gan.html
https://www.alexirpan.com/2017/02/22/wasserstein-gan.html

Labels:



Powered by Blogger