The squared-error of generalized LASSO: A precise analysis

Authors:

Abstract
We consider the problem of estimating an unknown but structured signal x 0 from its noisy linear observations y = Ax 0 + z ∈ ℝ m . To the structure of x 0 is associated a structure inducing convex function f(·). We assume that the entries of A are i.i.d. standard normal N(0, 1) and z ~ N(0, σ 2 I m ). As a measure of performance of an estimate x* of x 0 we consider the “Normalized Square Error” (NSE) ∥x* - x 0 ∥ 2 2 /σ 2 . For sufficiently small σ, we characterize the exact performance of two different versions of the well known LASSO algorithm. The first estimator is obtained by solving the problem argmin x ∥y - Ax∥ 2 + λf(x). As a function of λ, we identify three distinct regions of operation. Out of them, we argue that “R ON ” is the most interesting one. When λ ∈ R ON , we show that the NSE is D f (x 0 , λ)/m-D f (x 0 , λ) for small σ, where D f (x 0 , λ) is the expected squared-distance of an i.i.d. standard normal vector to the dilated subdifferential λ · ∂f(x 0 ). Secondly, we consider the more popular estimator argmin x 1/2∥y - Ax∥ 2 2 . + στ f(x). We propose a formula for the NSE of this estimator by establishing a suitable mapping between this and the previous estimator over the region R ON . As a useful side result, we find explicit formulae for the optimal estimation performance and the optimal penalty parameters λ* and τ*.

This publication has 25 references indexed in Scilit: