Wednesday January 29, 2003
Department of Mathematics, University of British Columbia
Abstract: The mammalian visual cortex contain multiple superimposed maps of different visual stimulus attributes, including a topographic map of each retina, plus locally smooth and periodic maps of edge orientation, spatial frequency, direction of motion and eye of origin. These maps can be referred to collectively as a polymap. I will show how many polymap properties can be reproduced by simple models, such as the Kohonen Self-organising Feature Map or Elastic Net algorithms. These models appear to work by maximising some combination of completeness and continuity in mapping from a high-dimensional stimulus space to a two-dimensional surface (the cortex). I will present evidence that real visual cortex maps optimise coverage. I will also discuss possible mappings from other plausible types of stimulus space and the limits that might apply to the number of different features that might be combined within a single polymap.
For more information contact J. Keener, 1-6089