• Regularized autoencoder
• Typically used to learn features for another task such as classification
• Loss function with penalty
• $$\mathcal{L}_{\text{sc}} = \underbrace{||WH - X||_2^2}_{\text{reconstruction term}} + \underbrace{\lambda ||H||_1}_{\text{sparsity term}}$$

• Generative Model
• Learns a latent variable model for its input data.
• Learning the parameters of a probability distribution modeling your data

Why VAE

Explore variations on data in a desired, specific direction

What the Variational Means