Refined least squares for support recovery

Sparse linear regression is a key step in many applications such as communication, image denoising or speech recognition. In these applications, the unknown signal of interest often modeled using set noisy equations. If vector coefficients (the signal) sparse, then its values can be identified even if system equations undetermined: recovery becomes possible with fewer observations than variables. this work, we study problem exact support based on and present Refined Least Squares (ReLS). Given measurements, our goal to recover (unknown) sparse signal. To use an average multiple least squares solutions, each computed subset full The estimated by identifying most significant solution. We demonstrate that wide variety settings method outperforms state-of-the-art algorithms.