JOURDAN Astrid

Enseignante-Chercheure

Member of team :

ICI

Lieu :

CYTech Pau, E203

Mail 1 :

Phone :

05 59 05 90 90

Biography

Astrid Jourdan is a teacher at CYTech (ex-EISTI) and researcher at ETIS Laboratory in CY Paris University. Head of the mathematics department of CYTech (2015-2018), and academic co-head of Pau campus. PhD in Applied Mathematics (2000), LMA laboratory, Pau university – HDR (2020), AGM Laboratory, CY Paris University.
Her research work focuses on the development of statistical methods to analyze CPU-expensive computer codes (sensitivity analysis methods, surrogate models, space-filling designs). Initially, her work was applied to oil industry simulators.  For the last 5 years, she develops these statistical tools to improve the convergence of metaheuristics and to interpret black-box machine learning models.
Co-PI of two European projects Eramus+ Capacity Building, TORUS, MONTUS, (2015-2022) on the statistical processing of data related to environmental sciences with Belgium, Italy, Vietnam, Cambodgia, Thailang.

Research activities

  • Construction of space-filling designs to determine which simulations to carry out in order to gather a maximum of information in a minimum of simulations. The interest is to reduce the combinatorial cost of a search organized according to a grid varying the input parameters of a model.
  • Sensitivity analysis which allows to determine which input variables have a real impact on the target variable (output). The objective is to reduce the dimension of a problem by eliminating the non-active variables and to better understand the relationship between the inputs and the output.
  • Construction of metamodels (kriging, SVM, neural networks,…) in order to replace a large computational code in different analyses (uncertainty quantification, optimization,…)

Ongoing projects

  • Co-PI of European Eramus+ Capacity building projet MONTUS. Restitution in November 2022.
  • Construction of space-filling designs for mixture experiments based on the Dirichlet distribution.
  • Integration of sensitivity analysis methods in the PSO metaheuristic.