23. Adaptive sampling and multi-fidelity surrogate modeling

Surrogate models are extracted from (expensive/time consuming) simulations. These surrogate models have different structures, ranging from simple interpolation techniques, local rational or polynomial methods, and even Gaussian Processes as used within Machine Learning. There are 2 important techniques that can be combined: by sampling the design space in an intelligent and adaptive way (adaptive sampling) and by combining (relatively fast) course simulations (with high simulation tolerances) with (slower) fine simulations (with low simulation tolerances).

The aim of this thesis is to determine the trade-off between accuracy and simulation time using adaptive sampling and the multi-fidelity surrogate modeling.


Back to top