An interpretable error indicator for neural-network surrogate models based on encoder–regression decomposition

Julien Yvonnet

doi:10.1016/j.cma.2026.119248

An interpretable error indicator for neural-network surrogate models based on encoder–regression decomposition

2026-08-11

0

PRE

AI

J

Julien Yvonnet

DOI：10.1016/j.cma.2026.119248

Original

Original request for help

Save

Abstract

En 中文

Neural-network-based surrogate models combined with dimensionality reduction techniques such as PCA/POD are widely used to approximate high-dimensional parametric systems. However, their deployment in engineering applications requires reliable error indicators capable of assessing prediction accuracy, including in regions not covered by the training data.

Keywords:

Neural networks

Error estimation

Surrogate models

Reduced-order models (ROMs)

Radial basis functions

Linear encoders

Real-time simulations

An interpretable error indicator for neural-network surrogate models based on encoder–regression decomposition

Abstract

Journal

Organization

Cited Papers

Citing Papers