Data-Efficient Sensitivity Analysis with Surrogate Modeling

Abstract

As performing many experiments and prototypes leads to a costly and long analysis process, scientists and engineers often rely on accurate simulators to reduce costs and improve efficiency. However, the computational demands of these simulators are also growing as their accuracy and complexity keeps increasing. Surrogate modeling is a powerful framework for data-efficient analysis of these simulators. A common use-case in engineering is sensitivity analysis to identify the importance of each of the inputs with regard to the output. In this work, we discuss surrogate modeling, sequential design, sensitivity analysis and how these three can be combined into a data-efficient sensitivity analysis method to accurately perform sensitivity analysis.