Speaker Name(s): Geoffrey Hinton, Ph.D., F.R.S.C.
Description: Recognizing a familiar shape is a difficult computational problem because foreground clutter may occlude large parts of the shape and the intensities of the pixels that remain are determined as much by the unknown viewpoint and lighting as by the shape. The brain performs this difficult computation very effectively by learning to extract multiple layers of features from the image. I will describe two different ways of learning features and I will show that a new model that uses the precise times of neural spikes to represent viewpoint information can explain a number of phenomena in mental imagery. This lecture is co-sponsored by NSERC and hosted by Ryerson University.