Thursday, April 20, 2000 B102 Wells Hall:

• 4:00 pm     Ingrid Daubechies
                   "Quantizing Redundant Representations"
• Abstract: Redundant representations are often superior to bases in order to extract features, and even to denoise signals. For compression purposes, the redundancy seems a handicap, however, mostly because we don't yet know how to "quantize" these expansions effectively. (Note: quantization stands here for something very different from the transition from classical mechanics to quantum mechanics - it is, rather, the transition from coefficients that can take a continuum of values to coefficients that are restricted to a discrete ("quantized") range.)
  
  For bandlimited signals, redundant representations have been used for a while in A/D conversion, via sigma-delta quantizers, which give an approach to quantize efficiently highly redundant samples. The talk will give an overview, from a mathamatical point of view, discuss some recent results, obtained jointly with Ron DeVore, and list open problems (of which there are many).

For additional information:

Department of Mathematics
Michigan State University
East Lansing, MI 48824-1027
(517)355-9680
ginther@math.msu.edu
http://www.math.msu.edu/Lecture_Series