Learning Latent Subspaces in Variational Autoencoders

Jack Klys, Jake Snell, Richard Zemel

Variational autoencoders (VAEs) are widely used deep generative models capable of learning unsupervised latent representations of data. Such representations are often difficult to…