Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ Online-Publikations-...arrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
https://dx.doi.org/10.25972/op...
Doctoral thesis . 2022
License: CC BY NC ND
Data sources: Datacite
versions View all 2 versions
addClaim

Proximal Methods for Nonconvex Composite Optimization Problems

Authors: Lechner, Theresa;

Proximal Methods for Nonconvex Composite Optimization Problems

Abstract

In Optimierungsproblemen mit zusammengesetzten Funktionen wird die Summe aus einer glatten und einer konvexen, nicht glatten Funktion minimiert. Die vorliegende Arbeit behan- delt mehrere numerische Verfahren zur Lösung solcher Probleme in endlich-dimensionalen Räumen, welche auf Proximity Operatoren basieren. Nach der Zusammenfassung einiger grundlegender Resultate aus der konvexen und nicht- glatten Analysis wird ein Verfahren erster Ordnung, das Proximal-Gradienten-Verfahren, vorgestellt und dessen Konvergenzeigenschaften ausführlich behandelt. Bekannte Resultate aus der Literatur werden dabei zusammengefasst und durch weitere Ergebnisse ergänzt. Im Anschluss werden im Hauptteil der Arbeit zwei Verfahren hergeleitet, die zusätzlich Informationen zweiter Ordnung nutzen und auf Proximal-Newton- beziehungsweise Proximal-Quasi- Newton-Verfahren beruhen. Der Unterschied zwischen beiden Verfahren liegt darin, dass bei ersterem eine klassische Schrittweitensuche verwendet wird, während das zweite stattdessen einen Regularisierungsparameter nutzt. Beide Techniken führen dazu, dass im Gegensatz zu vielen verwandten Verfahren in der jeweils ausführlichen Konvergenzanalyse die globale Konvergenz zu stationären Punkten ohne weitere einschränkende Voraussetzungen bewiesen werden kann. Ferner zeigen umfassende Resultate die lokalen Konvergenzeigenschaften sowie Konvergenzraten der Algorithmen auf, welche auf lediglich schwachen Annahmen beruhen. Ein Verfahren zur Lösung auftretender Proximal-Teilprobleme ist ebenfalls Bestandteil dieser Arbeit. Die Dissertation beinhaltet zudem eine umfangreiche Sammlung von Anwendungsbeispielen und zugehörigen numerischen Ergebnissen.

Optimization problems with composite functions deal with the minimization of the sum of a smooth function and a convex nonsmooth function. In this thesis several numerical methods for solving such problems in finite-dimensional spaces are discussed, which are based on proximity operators. After some basic results from convex and nonsmooth analysis are summarized, a first-order method, the proximal gradient method, is presented and its convergence properties are discussed in detail. Known results from the literature are summarized and supplemented by additional ones. Subsequently, the main part of the thesis is the derivation of two methods which, in addition, make use of second-order information and are based on proximal Newton and proximal quasi-Newton methods, respectively. The difference between the two methods is that the first one uses a classical line search, while the second one uses a regularization parameter instead. Both techniques lead to the advantage that, in contrast to many similar methods, in the respective detailed convergence analysis global convergence to stationary points can be proved without any restricting precondition. Furthermore, comprehensive results show the local convergence properties as well as convergence rates of these algorithms, which are based on rather weak assumptions. Also a method for the solution of the arising proximal subproblems is investigated. In addition, the thesis contains an extensive collection of application examples and a detailed discussion of the related numerical results.

Country
Germany
Related Organizations
Keywords

ddc:510, Optimierung, 510 Mathematik

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green