| Title: |
GORBIT: Global Optimization by Multistart Radial Basis Function Algorithms |
| Authors: |
Stefan Wild,
Christine Shoemaker |
| Abstract: |
We present a new algorithm, GORBIT, for global optimization when derivatives are unavailable. GORBIT uses a multistart procedure and a modified version of the local solver ORBIT to minimize the number of expensive objective function evaluations. We introduce a new multistart procedure, MIPE, that takes into account the function evaluations made by the local solver. The ORBIT solver uses local radial basis function models within trust-regions. ORBIT is modified in GORBIT to benefit from the function evaluations done in previous GORBIT iterations and to handle bound constraints. We present numerical results to determine the effect of the particular multistart procedure employed and to motivate the efficiency of GORBIT on the problem of finding large growth factors when doing Gaussian elimination. |
| Keywords: |
global optimization; derivative-free optimization; multistart methods; radial basis functions; blackbox optimization |
| Thanks: |
Stefan Wild was supported by a DOE Computational Science Graduate Fellowship under grant number DE-FG02-97ER25308, by an Argonne Director's Postdoctoral Fellowship, and by DOE under contract DE-AC02-06CH11357.
Christine Shoemaker was supported by NSF grants CCF-0305583 and DMS-0434390. |
| Status: |
Submitted, July 2009. |
| Link: |
[] |
| BibTeX: |
@techreport{SMWCAS09,
author = "Stefan M. Wild and Christine A. Shoemaker",
title = "GORBIT: Global Optimization by Multistart Radial Basis Function Algorithms",
institution = "Cornell University School of Operations Research and Information Engineering",
month = "July",
year = "2009",
number = "ORIE-1473"
}
|
Back to