Tackling box-constrained optimization via a new projected quasi-newton approach

Dongmin Kim, Suvrit Sra, Inderjit S. Dhillon

Publikation: Beitrag in FachzeitschriftArtikelBegutachtung

63 Zitate (Scopus)

Abstract

Numerous scientific applications across a variety of fields depend on box-constrained convex optimization. Box-constrained problems therefore continue to attract research interest. We address box-constrained (strictly convex) problems by deriving two new quasi-Newton algorithms. Our algorithms are positioned between the projected-gradient [J. B. Rosen, J. SIAM, 8 (1960), pp. 181-217] and projected-Newton [D. P. Bertsekas, SIAM J. Control Optim., 20 (1982), pp. 221- 246] methods. We also prove their convergence under a simple Armijo step-size rule. We provide experimental results for two particular box-constrained problems: nonnegative least squares (NNLS), and nonnegative Kullback-Leibler (NNKL) minimization. For both NNLS and NNKL our algorithms perform competitively as compared to well-established methods on medium-sized problems; for larger problems our approach frequently outperforms the competition.

OriginalspracheEnglisch
Seiten (von - bis)3548-3563
Seitenumfang16
FachzeitschriftSIAM Journal on Scientific Computing
Jahrgang32
Ausgabenummer6
DOIs
PublikationsstatusVeröffentlicht - 2010
Extern publiziertJa

Fingerprint

Untersuchen Sie die Forschungsthemen von „Tackling box-constrained optimization via a new projected quasi-newton approach“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren