Consistent parameter estimation for Lasso and approximate message passing (Q5894706)

The authors study the optimal tuning of the regularization parameter in Lasso (see [\textit{R. Tibshirani}, J. R. Stat. Soc., Ser. B 58, No. 1, 267--288 (1996; Zbl 0850.62538)] or [\textit{S. S. Chen} et al., SIAM J. Sci. Comput. 20, No. 1, 33--61 (1999; Zbl 0919.94002)]) or the threshold parameters in approximate message passing (AMP). They employ one of the standard model selection techniques, namely the Stein unbiased risk estimator (SURE) (see [\textit{C. M. Stein}, Ann. Stat. 9, 1135--1151 (1981; Zbl 0476.62035)]). For their goal, the authors use some important results given by \textit{M. Bayati} and \textit{A. Montanari} [IEEE Trans. Inf. Theory 57, No. 2, 764--785 (2011; Zbl 1366.94079)] and \textit{M. Bayati} and \textit{A. Montanari} [IEEE Trans. Inf. Theory 58, No. 4, 1997--2017 (2012; Zbl 1365.62196)]. Their main results are given in Theorem 3.3, Theorem 3.4 and Theorem 3.6. In fact, they show that the obtained estimates are consistent in the sense that they converge to their asymptotically optimal values in probability and proved asymptotic properties of the solution path of Lasso and AMP.