Skip to content

UQ Training Sept oct 2023

adutfoy edited this page May 12, 2023 · 1 revision

UQ Training : Methodology, OpenTURNS and Uranie, Sept-Oct 2023

This is the announce of training on High Performance Computing in Uncertainty Quantification, proposed by the OpenTURNS Consortium and CEA. This training was until last year part of the European Prace Program that stopped in december 2002.

Contacts

Cette formation propose de former les stagiaires à propager des incertitudes dans un code de calcul, selon une méthodologie en 4 étapes reconnue par la Communauté Incertitudes : spécification du problème, quantification des incertitudes, propagation, analyse de sensibilité.

L’étape de Spécification apprend au stagiaire à délimiter son problème d’incertitudes (incertitudes d’entrée, grandeur d’intérêt) et à en comprendre les caractéristiques (temps CPU d’évaluation de la fonction, gradients, …). L’étape de Quantification apprend à modéliser un vecteur de variables aléatoire par une loi de probabilité, définie par jugement d’expert ou par estimation à partir de données. L’étape de Propagation propage les incertitudes d’entrée jusqu’à la grandeur d’intérêt définie dans l’étape de Spécification. L’étape d’analyse de sensibilité permet de comprendre l’impact relatif des incertitudes d’entrée sur l’évaluation de la grandeur d’intérêt. La formation aborde aussi les aspects Optimisation et calibration de code, qui permet d’ajuster la loi d’une entrée aléatoire à partir de données d’une grandeur de sortie.

Les éléments théoriques présentés lors de la formation seront appliqués sur un cas jouet, à l’aide du Code OpenTURNS développé par le consortium OpenTURNS (EDF, Airbus, Onera, Phimeca, Imacs) ou du code Uranie développé par le CEA. Le langage utilisé est le Python.

La formation détaillera aussi les besoins en calcul haute performance du domaine particulier du traitement des incertitudes et développe les possibilités de calculs parallèles des plate-formes OpenTURNS et Uranie.

La formation est gratuite, ouverte à tous. Elle se déroule sur 3 jours, en ligne:

22, 29 septembre 2023 et 6 octobre 2023.

L’inscription est obligatoire à l’adresse :

https://indico.math.cnrs.fr/event/9777/

Si nécessaire, elle se déroule en langue anglaise.

Program

Day 1 : Methodology of Uncertainty Treatment - Basics of Probability and Statistics

  • General Uncertainty Methodology (30 min) : A. Dutfoy (G. Blondet)
  • Probability and Statistics: Basics (45 min) : G. Blondet (A. Dutfoy)
  • General introduction to OpenTURNS and Uranie (1h) : G. Blondet, JB Blanchard
  • Introduction to Python and Jupyter (45 min): practical work on distributions manipulations: G. Blondet (J. Pelamatti)

Lunch

  • Uncertainty Quantification (45min) : JB Blanchard (G. Blondet)
  • OpenTURNS - Uranie practical works: sections 1, 2 (1h): JB Blanchard , G. Blondet, (A. Dutfoy)
  • Central tendency and Sensitivity analysis (1h): A. Dutfoy (J. Pelamatti)

Day 2 : Quantification, Propagation and Ranking of Uncertainties

  • QCM on Day 1
  • Application to OpenTURNS and Uranie (1h): J. Pelamatti, G. Blondet, JB Blanchard
  • Estimation of probability of rare events (1h): G. Blondet (J. Pelamatti, JB Blanchard)
  • Application to OpenTURNS and Uranie (1h): J. Pelamatti, G. Blondet, JB Blanchard

Lunch

  • Distributed computing (1h) : Uranie (15 min, JB Blanchard), OpenTURNS (15 min, G. Blondet), Salome et OpenTURNS (30min, O. Mircescu)
  • Optimisation and Calibration (1h) : JB Blanchard , J. Pelamatti, (G. Blondet)
  • Application to OpenTURNS and Uranie (1h): JB Blanchard, J. Pelamatti, G. Blondet

Day 3 : HPC aspects - Meta model

  • QCM on Day 2
  • HPC aspects specific to the Uncertainty treatment - Example (1h) : K. Delamotte (J. Pelamatti)
  • Introduction to Meta models (validation, over-fitting) - Polynomial chaos expansion, Kriging meta model (2h) : J. Pelamatti

Lunch

  • Application to OpenTURNS and Uranie (1 ou 2h) : J. Pelamatti, G. Blondet, JB. Blanchard
  • Application industrielle : to be confirmed
  • Discussion / Participants projects