# GAN related things

This is a generative adversarial soft decision tree. Red regions imply $D$ thinks its from $p_x$, blue regions imply $D$ thinks its from $p_{G(z)}$.

This one is a linear WGAN-GP. I guess because of the gradient penalty $\lambda (\| \nabla_{\bar{x}}f(\bar{x})) \| - 1)^2$, the discriminator linear projector tries to make its gradient $1$ everywhere while trying to differentiate between fake and real.

This is a work in progress.

Written on January 27, 2019