Generative Models & Variational AutoEncoder Explained – Frank's World

The ever-increasing size of modern datasets combined with the difficulty of obtaining labeled information has made semi-supervised learning one of the problems of significant practical importance in modern data analysis. VAE offers a novel way to enforce structure on the representation surface, by doing so, it opens the possibility of employing traditional semi supervised learning techniques on the structured embedding space. In this talk, Shair Harel covers how VAE imposes latent space structure constraint, and how we can use it in a semi-supervised settings.