Wednesday, April 06, 2011

CS: The Dantzig Selector solvers: Multiplicative Noise Effect on the Donoho-Tanner Phase Transition (Part 8)

Salman Asif and I had a small discussion on the Homotopy Dantzig Selector solver he has developed. Looks like I was not using the right subroutines, so here are the new figures for the Donoho-Tanner phase transition.

if you recall the LASSO solver gave:


In other words, while similar at epsilon = 0.5/100, there is still a small advantage for LASSO in the low region of x (m/N). In other words, LASSO does "well" (compared to all other solvers) with 1% multiplicative noise in the under-determined regime.

No comments: