The paper is devoted to the estimation of small failure probabilities, i.e. on the reliability analysis, using the non-parametric model of the large-scale structure. Two major questions are addressed in this respect: (i) What are the effects of the model uncertainties on the prediction of very small failure probabilities, i.e. on the tails of the response distribution? This question will be answered by studying the differences in the failure probabilities predicted by the non-parametric approach and by the parametric approach, respectively. (ii) The non-parametric model of uncertainties features a very large number of random variables, thus leading to a high-dimensional reliability problem [3]. Due to the non-linear nature of the response, with respect to the space of the input random variables, this problem is very challenging. The performance of methods applicable to igh-dimensional reliability problems (Line Sampling, Subset Simulation) is studied.
