Explicit and Implicit Parametric Regression
David Barber (University College London)
Friday 18th October, 2013 15:00-16:00 Maths 204
I'll discuss some aspects of the recently popular deep belief/neural explicit networks and contrast their performance to implicit networks. These implicit networks can have excellent compression performance in images at a fraction of the computational effort require to train deep networks. As such I believe they deserve considerably more attention than has been given in the past. I'll discuss relations to convex bounding techniques for training and links to warped probabilistic mappings.