In this notebook we design and train a classic Auto-Encoder model on MNIST data set and learn its latent space using a conditional-GAN architecture. Then we visualize how the interpolation differs within AE's latent codes vs GAN's latent codes to support the idea of GAN better distribution grouping.