Artikel in Tagungsband INPROC-2016-12

Bibliograph.
Daten
Valentin, Julian; Pflüger, Dirk: Hierarchical Gradient-Based Optimization with B-Splines on Sparse Grids.
In: Garcke, Jochen (Hrsg); Pflüger, Dirk (Hrsg): Sparse Grids and Applications - Stuttgart 2014.
Universität Stuttgart, Fakultät Informatik, Elektrotechnik und Informationstechnik.
Lecture Notes in Computational Science and Engineering; 109, S. 315-336, englisch.
Springer, März 2016.
DOI: 10.1007/978-3-319-28262-6_13.
Artikel in Tagungsband (Konferenz-Beitrag).
CR-Klassif.G.1.6 (Numerical Analysis Optimization)
Keywordssparse grids; optimization; B-splines
Kurzfassung

Optimization algorithms typically perform a series of function evaluations to find an approximation of an optimal point of the objective function. Evaluations can be expensive, e.g., if they depend on the results of a complex simulation. When dealing with higher-dimensional functions, the curse of dimensionality increases the difficulty of the problem rapidly and prohibits a regular sampling. Instead of directly optimizing the objective function, we replace it with a sparse grid interpolant, saving valuable function evaluations. We generalize the standard piecewise linear basis to hierarchical B-splines, making the sparse grid surrogate smooth enough to enable gradient-based optimization methods. Also, we use an uncommon refinement criterion due to Novak and Ritter to generate an appropriate sparse grid adaptively. Finally, we evaluate the new method for various artificial and real-world examples.

Volltext und
andere Links
PDF (1379109 Bytes)
DOI
Abteilung(en)Universität Stuttgart, Institut für Parallele und Verteilte Systeme, Simulation großer Systeme
Eingabedatum18. April 2016
   Publ. Abteilung   Publ. Institut   Publ. Informatik