Date:	Wed, September 29, 2010
Time:	11:15
Place:	Research I Seminar Room

Abstract: Among various approaches in a compressed sensing paradigm, greedy algorithms have long established a place as a respectable competitor to ell-1 minimization. In this presentation we will introduce a way to measure performance of a deterministic greedy algorithm by comparing its error of approximation to the best possible, and discuss the classes of dictionaries and possible limitations on the signal that are of interest. In particular, we will present our most recent improvement in this area, namely, that the error of approximation by orthogonal greedy algorithm with regard to an M-coherent dictionary is within a constant factor of the best m-term approximation on essentially the full range of admissible m.