EUVIMEDEUVIMED entdecken
Crossref

Improving PINNs by Integrating Vanilla PINNs with Finite-Element-Informed Neural Networks

M. Mrini, A. Chahed, A. Bergam, A. El Harrak

Mathematical Modeling and Computing · 2026 · 13 (4) · 1055-1068

DOI: 10.23939/mmc2026.04.1055 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Mathematical Modeling and Computing, 2026. Verfasst von M. Mrini und 3 weiteren Autor:innen. Die Arbeit nennt 1 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Partial Differential Equations (PDEs) provide a fundamental mathematical tool for modeling a variety of physical and engineering phenomena. However, since analytical solutions of PDEs can be difficult and, in some cases, impossible to obtain, numerical methods have become the primary approach for solving PDEs. Recently, Deep Neural Networks (DNNs) have emerged as an alternative method for obtaining approximate solutions. Among these DNN-based approaches, Physics-Informed Neural Networks (PINNs) [1] have achieved significant advancements in solving differential equations. PINNs integrate physical knowledge into the neural network loss function through pointwise residual sampling. However, this strong-form definition of the loss function sometimes fails to produce sufficiently accurate approximate solutions. To enhance accuracy, we propose a hybrid approach that integrates both strong and weak forms inspired by the Finite Element Method [2] into the cost function definition, with the contribution of each part being adjusted automatically. To validate the performance of this approach, we conduct several experiments, including one-dimensional and two-dimensional equations.

Zitieren

APA

Mrini, M., Chahed, A., Bergam, A., & Harrak, A. E. (2026). Improving PINNs by Integrating Vanilla PINNs with Finite-Element-Informed Neural Networks. Mathematical Modeling and Computing, 13(4), 1055-1068. https://doi.org/10.23939/mmc2026.04.1055

Vancouver

Mrini M, Chahed A, Bergam A, Harrak AE. Improving PINNs by Integrating Vanilla PINNs with Finite-Element-Informed Neural Networks. Mathematical Modeling and Computing. 2026;13(4):1055-1068. doi:10.23939/mmc2026.04.1055

BibTeX
@article{Mrini2026,
  author = {Mrini, M. and Chahed, A. and Bergam, A. and Harrak, A. El},
  title = {Improving PINNs by Integrating Vanilla PINNs with Finite-Element-Informed Neural Networks},
  journal = {Mathematical Modeling and Computing},
  year = {2026},
  volume = {13},
  number = {4},
  pages = {1055--1068},
  doi = {10.23939/mmc2026.04.1055}
}
Kostenlos starten →