Submodular Maximization via Gradient Ascent: The Case of Deep Submodular Functions

We study the problem of maximizing deep submodular functions (DSFs) [13, 3] subject to a matroid constraint. DSFs are an expressive class of submodular functions that include, as strict subfamilies, the facility location, weighted coverage, and sums of concave composed with modular functions. We use...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Advances in neural information processing systems Jg. 2018; S. 7989
Hauptverfasser: Bai, Wenruo, Noble, William S, Bilmes, Jeff A
Format: Journal Article
Sprache:Englisch
Veröffentlicht: United States 01.12.2018
ISSN:1049-5258
Online-Zugang:Weitere Angaben
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:We study the problem of maximizing deep submodular functions (DSFs) [13, 3] subject to a matroid constraint. DSFs are an expressive class of submodular functions that include, as strict subfamilies, the facility location, weighted coverage, and sums of concave composed with modular functions. We use a strategy similar to the continuous greedy approach [6], but we show that the multilinear extension of any DSF has a natural and computationally attainable concave relaxation that we can optimize using gradient ascent. Our results show a guarantee of with a running time of ( ) plus time for pipage rounding [6] to recover a discrete solution, where is the rank of the matroid constraint. This bound is often better than the standard 1 - 1 guarantee of the continuous greedy algorithm, but runs much faster. Our bound also holds even for fully curved ( = 1) functions where the guarantee of 1 - degenerates to 1 - 1 where is the curvature of [37]. We perform computational experiments that support our theoretical results.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:1049-5258