Inspired by the approach of this paper and our own experiments, we optimize a complicated function as follows:
1. Get an initial estimate
2. Generate M perturbed variants of the inital estimate by adding noise
3. Evaluate the cost of all M starting points
4. Optimize over the N < M best solutions
5. Select the one with the lowest cost as final estimate
Is there some "official" name for such an approach?
The closest I could find is repeated local search but this does not include the N < M step.
