October 29, 2018
@hardmaru
scott mccloud tom white, sampling generative networks (2016)
BEGAN ( a GAN that works like a VAE, in that you can interpolate all the latent vectors )
(x, y, P-touch, P-lift, P-end) as a row of data
RNN -> latent vector -> add gaussian noise to latent vector -> decode RNN
World models
echo state networks?
otoro.net
creatures avoiding planks
worldmodels.github designrl.github