Summary: 
We prove that nonnegative least squares (typically prone to over-fitting) can be slightly modified to return sparse results.
Abstract: 

This letter demonstrates that sparse recovery can be achieved by an L1-minimization ersatz easily implemented using a conventional nonnegative least squares algorithm. A connection with orthogonal matching pursuit is also highlighted. The preliminary results call for more investigations on the potential of the method and on its relations to classical sparse recovery algorithms.

Authors: 
Simon Foucart and David Koslicki
Journal: 
IEEE Signal Processing Letters
Date: 
Thursday, February 27, 2014
Article link: