EGO
Efficient Global Optimization: kriging surrogates and expected improvement
Paper title Efficient Global Optimization of Expensive Black-Box Functions
Paper doi.org
Fits a kriging (Gaussian-process) response surface to a space-filling Latin hypercube design of about ten runs per variable, checks it by cross-validation, then picks each next run where the expected improvement over the best design so far is largest, found with a branch-and-bound search. Expected improvement weighs a low predicted value against a large predicted error, balancing exploitation of the surface with improving it where it is unsure; the search stops when expected improvement falls below 1% of the best value.
- Published
- Journal of Global Optimization
Share on LinkedIn, X or Bluesky, or or .
Know related work that belongs here? Suggest a link
Something wrong on this page? Tell the editor