Tag
The paper resolves contradictions in using neural networks for derivative-free optimization by identifying three key factors—role, radius, and room—that determine when learned local models improve performance in simulation optimization.