Surrogate Modeling for Stochastic Simulators: An Overview and Recent Developments

Xujia Zhu (CentraleSupélec, L2S)
March 08, 2024 — 11:00 — "new L2S location (IBM building), 3rd floor room" (and Teams)

Abstract

In the context of uncertainty quantification or optimization, it is indispensable to evaluate computational models repeatedly. This task is intractable for expensive numerical models due to prohibitively high computational cost. The challenge intensifies when dealing with stochastic simulators, as several model evaluations with the same input parameters would yield different values of the model response. Given the intrinsic stochasticity in the output, the direct application of classical surrogate models to emulate stochastic simulators is not viable. In this talk, I will provide a comprehensive overview of the development of stochastic surrogate models. In the second part, I will present two surrogate models, namely the generalized lambda model and stochastic polynomial chaos expansions, developed through our recent years of research to emulate the complete response distribution of stochastic simulators.

Bio

Xujia Zhu is currently an assistant professor at CentraleSupélec, Paris-Saclay University, and is affiliated with the Laboratory of Signals and Systems. His primary research focuses on the confluence of numerical simulations and statistics, encompassing a diverse array of subjects surrounding uncertainty quantification, including surrogate modeling, uncertainty propagation, sensitivity analysis, and reliability analysis. Prior to joining CentraleSupélec, he obtained an engineer’s degree in mechanics from the École Polytechnique (France) in 2015. Subsequently, in 2017, he received a master’s degree (with high distinction) in computational mechanics from the Technical University of Munich (Germany). In 2022, he completed his Ph.D. from the Chair of Risk, Safety, and Uncertainty Quantification at ETH Zurich (Switzerland). Then, he continued his academic journey as a postdoctoral researcher at the same institution until the end of 2023. Drawing upon his multidisciplinary background, he has collaborated with engineers and researchers across various fields, among others, civil engineering, agriculture, and biology.

Reference

Xujia Zhu (2023). Surrogate modeling for stochastic simulators using statistical approaches. Ph.D. thesis, ETH Zurich. https://doi.org/10.3929/ethz-b-000604116.