Simple algorithms for sparse linear regression with uncertain covariates

Yudong Chen, Constantine Caramanis

Abstract: In this short paper we consider sparse linear regression with missing or noisy covariates. This problem has recently attracted some attention, with the best known results given in [1], where they use a projected gradient approach to approximately solve a non convex optimization problem. Here we show that an extremely simple, lower complexity algorithm that achieves the same (or better) bounds for support recovery.