COMPUTATION OF SOBOL INDICES IN GLOBAL SENSITIVITY ANALYSIS FROM SMALL DATA SETS BY PROBABILISTIC LEARNING ON MANIFOLDS

Global sensitivity analysis provides insight into how sources of uncertainty contribute to in predictions computational models. indices, also called variance-based indices and Sobol are most often computed with Monte Carlo methods. However, when the model is computationally expensive only a small number samples can be generated, that is, so-called small-data settings, usual estimates may lack sufficient accuracy. As means improving accuracy such we explore use probabilistic learning. The objective learning learn from available used generate additional samples, which global then deduced. We demonstrate interest method by applying it an illustration concerned forecasting contribution Antarctic ice sheet sea level rise.