Generative model that learns a latent distribution by optimizing a variational lower bound, useful for synthesis and representation learning.