BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Recent Developments in Surrogate Modeling for Stochastic Simulator
 s: Comprehensive Overview and Insights - Xujia Zhu (CentraleSupélec)
DTSTART:20250827T133000Z
DTEND:20250827T140000Z
UID:TALK234499@talks.cam.ac.uk
DESCRIPTION:Over the past few decades\, surrogate models\, also known as m
 etamodels or emulators\, have emerged as essential tools for enabling effi
 cient uncertainty quantification in complex computational systems. These m
 odels provide fast approximations of expensive simulations\, making them c
 ritical in settings where large numbers of model evaluations are needed\, 
 such as uncertainty propagation\, reliability analysis\, and sensitivity a
 nalysis.&nbsp\;\nMuch of the methodological development in surrogate model
 ing has focused on deterministic simulators\, where a given set of input p
 arameters produces a single\, repeatable output value. In contrast\, many 
 real-world simulators incorporate intrinsic stochasticity\, producing diff
 erent output values across repeated evaluations at the same input. Such st
 ochastic simulators arise in diverse applications\, including those involv
 ing stochastic processes\, agent-based models\, and simulations incorporat
 ing experimental outcomes. The inherent randomness in these systems requir
 es a fundamental shift in surrogate modeling strategies\, as traditional m
 ethods for deterministic models\, such as Gaussian process regression or p
 olynomial chaos expansions\, do not adequately capture the output variabil
 ity or distributional structure.\nThis talk will provide a comprehensive o
 verview of the state of the art in surrogate modeling for stochastic simul
 ators\, covering key conceptual distinctions\, modeling objectives\, and r
 epresentative approaches in statistics and machine learning. Among the met
 hods reviewed\, particular attention will be given to two surrogate modeli
 ng approaches developed to emulate the full output distribution\, with pri
 mary applications in engineering: the generalized lambda model\, which fle
 xibly captures response distributions through parametric families\, and st
 ochastic polynomial chaos expansions\, which represent output randomness t
 hrough additional latent variables. These methods offer promising directio
 ns for constructing efficient surrogates in the presence of intrinsic stoc
 hasticity. The talk will conclude with a discussion of open challenges and
  future directions.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR