To understand the role that GPs can play in optimizing a blackbox function,
- i.e., one about which one knows little (it is opaque to the optimizer)
- and which can only be probed through expensive evaluation.
Basically, the idea is to view optimization as an application of sequential design.
The role of "modeling" in optimization has a rich history,
- and we'll barely scratch the surface there.
But the potential role of modern statistical modeling is just recently being realized by the mathematical programming, statistics, and machine learning communities.