Computing Statistical Moments Via Tensorization of Polynomial Chaos Expansions

We present an algorithm for estimating higher-order statistical moments of multidimensional functions expressed as polynomial chaos expansions (PCE). The algorithm starts by decomposing the PCE into a low-rank tensor network using a combination of tensor-train and Tucker decompositions. It then effi...

Descripción completa

Detalles Bibliográficos
Autor: Ballester Ripoll, Rafael
Tipo de recurso: artículo
Fecha de publicación:2024
País:España
Institución:IE
Repositorio:Repositorio IE
OAI Identifier:oai:repositorio.ie.edu:20.500.14417/4021
Acceso en línea:https://doi.org/10.1137/23M155428X
https://hdl.handle.net/20.500.14417/4021
https://epubs.siam.org/doi/10.1137/23M155428X
Access Level:acceso abierto
Palabra clave:33 Ciencias Tecnológicas
ODS 9 - Industria, innovación e infraestructura
polynomial chaos expansions
statistical moments
surrogate modeling
tensor decompositions
tensor train decomposition
Tucker decomposition
Descripción
Sumario:We present an algorithm for estimating higher-order statistical moments of multidimensional functions expressed as polynomial chaos expansions (PCE). The algorithm starts by decomposing the PCE into a low-rank tensor network using a combination of tensor-train and Tucker decompositions. It then efficiently calculates the desired moments in the compressed tensor domain, leveraging the highly linear structure of the network. Using three benchmark engineering functions, we demonstrate that our approach offers substantial speed improvements over alternative algorithms while maintaining a minimal and adjustable approximation error. Additionally, our method can calculate moments even when the input variable distribution is altered, incurring only a small additional computational cost and without requiring retraining of the regressor.