Representation Learning with Latent Variable Models

In this post, I discuss what makes the posterior distribution of a directed latent variable model a useful representation. Also, I raise questions that deserve careful study in the case of undirected latent variable models.