“…Consequently, MC methods are often computationally intractable for high‐fidelity simulation models because they require a prohibitively large number of evaluations to obtain moderate accuracy in response statistics. In contrast, surrogate methods such as polynomial chaos, Gaussian processes, low‐rank decompositions, and sparse grid interpolation can be used to build an approximation of the input‐output response, often at a fraction of the cost of MC sampling. Once the surrogate has been constructed, various uncertainty quantification (UQ) tasks, such as sensitivity analysis, density estimation, etc., can then be performed on the approximation at negligible cost *…”