Recent Developments in Surrogate Modeling for Stochastic Simulators: Comprehensive Overview and Insights

Add to your list(s) Download to your calendar using vCal

• Xujia Zhu (CentraleSupélec)
• Wednesday 27 August 2025, 14:30-15:00
• Seminar Room 1, Newton Institute.

If you have a question about this talk, please contact nobody.

RCLW04 - Early Career Pioneers in Uncertainty Quantification and AI for Science

Over the past few decades, surrogate models, also known as metamodels or emulators, have emerged as essential tools for enabling efficient uncertainty quantification in complex computational systems. These models provide fast approximations of expensive simulations, making them critical in settings where large numbers of model evaluations are needed, such as uncertainty propagation, reliability analysis, and sensitivity analysis.  Much of the methodological development in surrogate modeling has focused on deterministic simulators, where a given set of input parameters produces a single, repeatable output value. In contrast, many real-world simulators incorporate intrinsic stochasticity, producing different output values across repeated evaluations at the same input. Such stochastic simulators arise in diverse applications, including those involving stochastic processes, agent-based models, and simulations incorporating experimental outcomes. The inherent randomness in these systems requires a fundamental shift in surrogate modeling strategies, as traditional methods for deterministic models, such as Gaussian process regression or polynomial chaos expansions, do not adequately capture the output variability or distributional structure. This talk will provide a comprehensive overview of the state of the art in surrogate modeling for stochastic simulators, covering key conceptual distinctions, modeling objectives, and representative approaches in statistics and machine learning. Among the methods reviewed, particular attention will be given to two surrogate modeling approaches developed to emulate the full output distribution, with primary applications in engineering: the generalized lambda model, which flexibly captures response distributions through parametric families, and stochastic polynomial chaos expansions, which represent output randomness through additional latent variables. These methods offer promising directions for constructing efficient surrogates in the presence of intrinsic stochasticity. The talk will conclude with a discussion of open challenges and future directions.

This talk is part of the Isaac Newton Institute Seminar Series series.

This talk is included in these lists:

• All CMS events
• Featured lists
• INI info aggregator
• Isaac Newton Institute Seminar Series
• School of Physical Sciences
• Seminar Room 1, Newton Institute
• bld31

Note that ex-directory lists are not shown.