Noisy Optimization
Abstract
The black box complexity of noisy-optimization is a great research area, with many real-world applications. Various criteria can be used: cumulative regret, simple regret, uniform rates. We discuss the impact of the use of second order information (improved rates under low noise assumption), or local sampling (slower simple regret convergence), or evolutionary optimization with revaluations (as efficient as mathematical programming in some cases with cumulative regret).
Origin : Files produced by the author(s)