Extending multi-fidelity surrogate modelling to stochastic simulators


Loading...

Date

2024-02

Publication Type

Other Conference Item

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

Multi-fidelity surrogate models (MFSMs) fuse information from models with varying computational fidelities into a new surrogate model (SM). Such models can predict the output of complex systems with a higher accuracy and lower cost compared to single-fidelity SMs. MFSMs can be constructed based on high- and lower-fidelity data from physical experiments and/or computer simulations. In real-world applications, experiments are often contaminated by measurement noise, while computer simulations may be polluted by numerical noise. This noise introduces a non-deterministic element, rendering the response of a model or experiment a random variable, even when the input parameters are fixed. This non-deterministic model behaviour is typical of stochastic simulators. These simulators are pertinent when latent sources of uncertainty impact a system, resulting in the response being a random variable conditioned on the input parameters. Stochastic simulators can effectively capture the noisy nature of both high- and lower-fidelity models. To alleviate the cost of repeatedly evaluating stochastic simulators, different methods exist to emulate their response. Such approaches can be statistical, replication-, or spectral expansions-based. This work explores the synergy between MFSMs and stochastic simulators, paving the way for the development of multi-fidelity stochastic emulators.

Publication status

published

External links

Editor

Book title

Journal / series

Volume

Pages / Article No.

Publisher

ETH Zurich, Risk, Safety and Uncertainty Quantification

Event

SIAM Conference on Uncertainty Quantification (UQ 2024)

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Organisational unit

03962 - Sudret, Bruno / Sudret, Bruno check_circle

Notes

Conference lecture held on February 28, 2024.

Funding

955393 - European Training Network on Grey-Box Models for Safe and Reliable Intelligent Mobility Systems (EC)

Related publications and datasets