Distinguished Lecture Series | Prof. Bruno Sudret | 2025-11-05 скачать в хорошем качестве

Distinguished Lecture Series | Prof. Bruno Sudret | 2025-11-05

TITLE: Surrogate modelling for stochastic simulators by Prof. Bruno Sudret, Chair of Risk, Safety and Uncertainty Quantification, ETH Zurich, Switzerland ABSTRACT: Computational models, or simulators, play a central role in engineering and applied sciences for designing and assessing complex systems. However, advanced analyses such as optimization and uncertainty quantification typically require a large number of model evaluations, making brute-force approaches like Monte Carlo simulation computationally prohibitive. This has motivated the development of surrogate models, such as polynomial chaos expansions and Gaussian process emulators, that provide efficient approximations of costly simulators. In many modern applications, including epidemiology, mathematical finance, and wind turbine design, the underlying simulators are stochastic: even for fixed input parameters, repeated runs yield different outputs due to internal sources of randomness. Consequently, each input realization corresponds to a random response, which must be characterized by its conditional distribution. The talk presents recent advances in surrogate modelling for stochastic simulators, called stochastic emulators. The generalized lambda model (GLaM) will be introduced. It combines parametric lambda distributions with polynomial chaos expansions to emulate full output distributions without requiring replicated simulations. Bruno then discusses stochastic polynomial chaos expansions, which can capture more complex, including multimodal, response distributions. Finally, he outlines a spectral approach based on random field representations for cases where full simulator trajectories are available. The methods will be illustrated with examples from engineering applications. The software package UQLab, which contains a module for stochastic emulators, will also be presented. Bruno has been a Professor of Risk, Safety and Uncertainty Quantification (UQ) at ETH Zurich since 2012. His research and teaching focus is on computational methods for uncertainty quantification, surrogate modelling, reliability and sensitivity analysis, Bayesian model calibration, and reliability-based design optimization. He graduated from École Polytechnique (France) in 1993 and earned both his MSc and PhD in Civil Engineering from École Nationale des Ponts et Chaussées (France) in 1996 and 1999, respectively. After a postdoctoral appointment at the University of California, Berkeley, he joined Electricité de France R&D, where he led a group in probabilistic engineering mechanics. Afterwards he served as Director of Research and Strategy at Phimeca Engineering (France). Bruno has (co-)authored more than 250 journal and conference papers. He serves on the editorial boards of Reliability Engineering & System Safety, Probabilistic Engineering Mechanics, and Structural Safety. He also leads the development of UQLab, a general-purpose software platform for uncertainty quantification, and the community platform UQWorld, dedicated to disseminating UQ methods and best practices. LICENSE: CC BY-SA 4.0 CASUS – Center for Advanced Systems Understanding Helmholtz-Zentrum Dresden-Rossendorf e.V. (HZDR) Untermarkt 20 D-02826 Görlitz