CPL - Chalmers Publication Library
| Utbildning | Forskning | Styrkeområden | Om Chalmers | In English In English Ej inloggad.

Fast Candidate Points Selection in the LASSO Path

Ashkan Panahi (Institutionen för signaler och system, Signalbehandling) ; Mats Viberg (Institutionen för signaler och system, Signalbehandling)
Ieee Signal Processing Letters (1070-9908). Vol. 19 (2012), 2, p. 79-82.
[Artikel, refereegranskad vetenskaplig]

The LASSO sparse regression method has recently received attention in a variety of applications from image compression techniques to parameter estimation problems. This paper addresses the problem of regularization parameter selection in this method in a general case of complex-valued regressors and bases. Generally, this parameter controls the degree of sparsity or equivalently, the estimated model order. However, with the same sparsity/model order, the smallest regularization parameter is desired. We relate such points to the nonsmooth points in the path of LASSO solutions and give an analytical expression for them. Then, we introduce a numerically fast method of approximating the desired points by a recursive algorithm. The procedure decreases the necessary number of solutions of the LASSO problem dramatically, which is an important issue due to the polynomial computational cost of the convex optimization techniques. We illustrate our method in the context of DOA estimation.

Nyckelord: Homotopy, LARS, LASSO, linear regression, stagewise regression, regression, model

Denna post skapades 2012-02-07. Senast ändrad 2015-05-08.
CPL Pubid: 154817


Läs direkt!

Lokal fulltext (fritt tillgänglig)

Länk till annan sajt (kan kräva inloggning)

Institutioner (Chalmers)

Institutionen för signaler och system, Signalbehandling (1900-2017)



Chalmers infrastruktur

Relaterade publikationer

Denna publikation ingår i:

Parameter Estimation Using Sparse Modeling: Algorithms and Performance Analysis