STELA algorithm for sparsity regularized linear regression (LASSO)

STELA algorithm solves the following optimization problem:

It is based on the parallel best-response (Jacobi) algorithm with guaranteed convergence. It exhibits a fast, reliable and stable performance.

Reference: Sec. IV-C of Y. Yang, and M. Pesavento, "A unified successive pseudoconvex approximation framework", IEEE Transactions on Signal Processing, vol. 65, no. 13, pp. 3313-3328, Jul. 2017. URL: IEEE, Arxiv

Input Parameters:

A :      N * K matrix,  dictionary;

y :      K * 1 vector,  noisy observation;

mu:      scalar, regularization gain;

MaxIter: maximum number of iterations, default = 1000;

Definitions:

Output Parameters:

objval: objective function value (f + g);

x:      K * 1 vector, the optimal variable

error:  specifies the solution precision (a smaller error implies a better solution);