How do we learn invariant representations?

Yann LeCun , The Courant Institute of Mathematical Sciences, New York, NY.

A major challenge for machine learning is to devise methods that can learn appropriate features and internal representations from data, labeled and unlabeled. But how can machines learn representations of the perceptual world that are robust to irrelevant variations of the inputs? I will describe a few attempts at designing learning algorithms that can learn hierarchies of invariant features and various applications, notably in musical genre recognition, and visual scene parsing. Read more ...

Published on : Wednesday 13 June 2012