NOVEL PARADIGM FOR APPROXIMATING SYSTEMS USING SURROGATE MODELS

Abstract

Complex systems are typically studied via computer experiments which involve experimenting on a rigorous first-principles model instead of a real system. Revolutionary advances in computing technologies over the last two decades have empowered researchers to incorporate greater details into such models. However, this comes at the expense of larger model size and greater computational burden. Therefore, it is beneficial to replace high-fidelity models by computationally cheaper surrogate models that offer a simpler overall picture. With this motivation, our research focuses on two aspects of surrogate construction viz. sampling and surrogate selection that has resulted in two novel approaches namely SSA and LEAPS2. SSA is a novel adaptive sampling approach that combines both spatial and quality aspects to achieve optimal sample placement. LEAPS2 is a learning-based evolutionary paradigm for selecting the best surrogate. Overall, this thesis presents a theoretical development of these two paradigms and highlights their practicality via extensive evaluation.