Deep Generative Latent Models
In machine learning, a latent variable model attempts to capture the belief that real world possibly high dimensional data is generated by some low dimensional latent variables. For example, real world data in the form of pixels of hand written digits may have low dimensional structure in which digit, thickness of the digit, curviness, etc. Generative refers to the ability to generate new artificial instances of data. Deep refers to the methodology of using deep neural networks to do this modeling. This talk will be an introduction to these concepts and the methodology involved. We will also touch on very preliminary results in learning invariant or fair latent models.