Predavanje naslova Parallel surrogate optimization of building energy models dr. sc. Sanje Stevanović s Mathematical Institute of the SANU (Belgrade), održat će se u četvrtak, 11. lipnja 2026., u 13 sati u predavaonici 121 na Građevinskom fakultetu.

Sažetak predavanja nalazi se u nastavku.

Surrogate optimization addresses expensive black-box optimization problems by replacing the true objective with a cheaper surrogate model that guides the selection of new evaluation points. The lecture will focus on our recent surrogate optimization approach for mixed design spaces with continuous, integer, and categorical variables [1]. We will outline how these methods combine machine-learning surrogates, uncertainty estimates, adaptive exploration–exploitation balance, and parallel evaluations to efficiently search for optima. As a motivating application, we will show how surrogate optimization can be used in simulation-based building energy optimization, where tools such as EnergyPlus make each function evaluation costly.