This is not a programming problem, it is about a mathematical algorithm.
Quote:
I have an N-dimensional distribution that resembles the Dirac Delta Function (there is a single narrow peak in an otherwise uniform distribution). My sampling cost is very high, so I want to minimize the number of samples required to locate the peak.
There is no magic, if you get useful information only close to the narrow peak, only systematic sampling will get the solution.
The only hope is to find an algorithm that will cover the field of possible solutions with minimum sampling.
My algorithm would probably look like:
I think I would consider the field as a grid of 1 square and sample each corner.
If nothing found, split each square in 4 and repeat until a useful sampling is found.
When on a useful sampling repeat the algorithm locally.