Invariance Learning

Peter Foldiak

Psychological Laboratory,
University of St. Andrews,
St. Andrews, U.K.

We see images but we are interested in objects. The relationship between an image, and its real causes are usually complex, and trivial changes in the viewing conditions can cause significant changes in most simple measures of the image. We need to learn to connect different representations of the same object, and the temporal coherence of the natural environment helps to solve this problem. A neural network learning rule exploiting temporal coherence may also be useful for learning coordinate transformations for the neural representations of space.

The lecture will take place:

in the Lecture Hall, Room 203, 44 Cummington St.
on Tuesday, July 16, 2002
at 1:00 pm