Lecture: Course 9.S912 "Why some deep learning architectures work: the computational magic of the ventral stream"

SpeakerTomaso Poggio
AffiliationCBCL, McGovern Institute, BCS Dept., CSAIL, MIT
Date and Time Sept. 13, 2013, 11 a.m. - 02:00 p.m.
LocationMIT Bldg 46-3189, McGovern 3rd Floor Seminar Room
HostTomaso Poggio

Tomaso Poggio will be speaking this Friday in 9.S912 (see abstract below). If you would like to receive speaker announcements please subscribe to the mailing list at: http://mailman.mit.edu/mailman/listinfo/vision_and_learning

Abstract: I will explain a recent, unpublished theory of hierarchical architectures and,in particular, of the ventral stream in visual cortex. The initial assumption is that the computational goal of the ventral stream is to compute a representation of objects which is invariant to transformations. The theory shows how a process based on high-dimensional dot products can use stored "movies" of objects transforming, to encode new images in an invariant way. Theorems show that invariance implies several properties of the ventral stream organization and of the tuning of its neurons. The main contribution is a theoretical framework for the next phase of machine learning beyond supervised learning: the unsupervised learning of representations that reduce the sample complexity of the final supervised learning stage.

There are no comments yet

Post a new comment...