Capturing Provenance to Improve the Model Training of PINNs: first handon experiences with Grid5000 - Archive ouverte HAL Access content directly
Conference Papers Year : 2021

Capturing Provenance to Improve the Model Training of PINNs: first handon experiences with Grid5000

(1) , (1) , (1) , (2) , (3) , (1) , (1)
1
2
3
Rômulo M. Silva
  • Function : Author
  • PersonId : 1118054
Débora Pina
  • Function : Author
  • PersonId : 1079404
Liliane Kunstmann
  • Function : Author
  • PersonId : 1118048
Patrick Valduriez
Marta Mattoso
  • Function : Author
  • PersonId : 863196

Abstract

The growing popularity of Neural Networks in computational science and engineering raises several challenges in configuring training parameters and validating the models. Machine learning has been used to approximate costly computational problems in computational mechanics, discover equations by coefficient estimation, and build surrogates. Those applications are outside of the common usage of neural networks. They require a different set of techniques generally encompassed by Physics-Informed Neural Networks (PINNs), which appear to be a good alternative for solving forward and inverse problems governed by PDEs in a small data regime, especially when it comes to Uncertainty Quantification. PINNs have been successfully applied for solving problems in fluid dynamics, inference of hydraulic conductivity, velocity inversion, phase separation, and many others. Nevertheless, we still need to investigate its computational aspects, especially its scalability, when running in large-scale systems. Several hyperparameter configurations have to be evaluated to reach a trained model, often requiring fine-tuning hyperparameters, despite the existence of a few setting recommendations. In PINNs, this fine-tuning requires analyzing configurations and how they relate to the loss function evaluation. We propose provenance data capture and analysis techniques to improve the model training of PINNs. We also report our first experiences on running PINNs in Grid5000 using hybrid CPU-GPU computing.
Fichier principal
Vignette du fichier
CILAMCE-2021-Authors.pdf (619.18 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

lirmm-03443548 , version 1 (23-11-2021)

Identifiers

  • HAL Id : lirmm-03443548 , version 1

Cite

Rômulo M. Silva, Débora Pina, Liliane Kunstmann, Daniel de Oliveira, Patrick Valduriez, et al.. Capturing Provenance to Improve the Model Training of PINNs: first handon experiences with Grid5000. CILAMCE-PANACM 2021 - XLII Ibero-Latin-American Congress on Computational Methods in Engineering - III Pan-American Congress on Computational Mechanics, Nov 2021, Rio de Janeiro, Brazil. pp.1-7. ⟨lirmm-03443548⟩
58 View
75 Download

Share

Gmail Facebook Twitter LinkedIn More