# Unsupervised learning Lots of potential, but lacking ideas and breakthroughs. - Variational Altering Code with reparameterization tricks - Generative adversarial nets - One of the biggest ideas - Sparsity and slow features - Geoffrey Hinton thinks slow features are a mistake - He thinks that we shouldn't go for features that don't change, but rather for features that change in a predictable way Basic principle shared by Geoffrey Hinton for modeling anything: - Take your measurements - Apply nonlinear transformations to your measurements until you get to a representation as a state vector, in which the action is linear - You don't just pretend it's linear like you do with common filters - Instead, you actually find a transformation from the observables to the underlying variables, where linear operations like matrix multipliers on the underlying variables will do the work - For example if you want produce an image based on another one from another viewpoint - Go from the pixels to coordinates - Once you have the coordinate representation, do a matrix multiplier to change viewpoint - Map it back to pixels