/variational-auto-encoder

Variational auto-encoder in MNIST images

Primary LanguageJupyter Notebook

varitional-auto-encoder

Variational auto-encoder in MNIST images

  • Using KL divergence in auto encoder to make it generative
  • Estimating distributions of the latent variable for a image rather then solid points
  • Thus the network not only memorizes the latent variable of a image to decode but rather learns the features of the images
  • First step in learning geanrative networks

Some images generated by the VAE which had 2 latent dimension

2D VAE Sprite

3-D representation of the latent variable according to the digit it represents in vae with 3 latent spaces

Latent

Playing around with latent variables to construct new digits

  • Here we can see that new generated images carry some property of digits

test test test