Surrogate Modeling

Surrogate modeling aims to create computationally inexpensive approximations of complex simulations, primarily to accelerate design optimization and uncertainty quantification in various scientific and engineering domains. Current research emphasizes developing accurate and efficient surrogate models using diverse architectures, including neural networks (e.g., DeepONets, U-Nets, recurrent neural networks), Gaussian processes, polynomial chaos expansions, and large language models, often incorporating physics-based constraints or multi-fidelity approaches to improve performance. This field is significant because it enables faster exploration of design spaces, more robust uncertainty analysis, and improved decision-making in applications ranging from fluid dynamics and reservoir simulation to materials science and nuclear engineering.