EUVIMEDEUVIMED entdecken
Crossref

Deep neural networks as discrete dynamical systems: Implications for physics-informed learning

Abhisek Ganguly iD, Santosh Ansumali iD, Sauro Succi iD

The Journal of Chemical Physics · 07.07.2026 · 165 (1)

DOI: 10.1063/5.0315971 ↗

Kurz gefasst

Fachartikel in der Zeitschrift The Journal of Chemical Physics, online erschienen am 01.07.2026. Verfasst von Abhisek Ganguly und 2 weiteren Autor:innen. Beteiligte Einrichtung: Engineering Mechanics Unit, Jawaharlal Nehru Centre for Advanced Scientific Research, Jakkur, Bangalore 560064, Karnataka, India. Die Arbeit nennt 33 Literaturangaben. Gefördert von: Scuola Internazionale Superiore di Studi Avanzati, HORIZON EUROPE European Innovation Council, Abdul Kalam Technology Innovation National Fellowship.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We revisit the analogy between feed-forward deep neural networks (DNNs) and discrete dynamical systems derived from neural integral equations and their corresponding partial differential equation forms. A comparative analysis between the numerical/exact solutions of the Burgers’ and Eikonal equations and those obtained via physics-informed neural networks (PINNs) is presented. We show that PINN learning provides a different computational pathway compared to standard numerical discretization in approximating essentially the same underlying dynamics of the system. Within this framework, DNNs can be interpreted as discrete dynamical systems whose layerwise evolution approaches attractors, and multiple parameter configurations may yield comparable solutions, reflecting the degeneracy of the inverse mapping. In contrast to the structured operators associated with finite-difference procedures, PINNs learn dense parameter representations that are not directly associated with classical discretization stencils. This distributed representation generally involves a larger number of parameters, leading to reduced interpretability and increased computational cost. However, the additional flexibility of such representations may offer advantages in high-dimensional settings where classical grid-based methods become impractical.

Autor:innen und Institutionen

  • Engineering Mechanics Unit, Jawaharlal Nehru Centre for Advanced Scientific Research, Jakkur, Bangalore 560064, Karnataka, India.
  • Italian Institute of Technology, Viale Regina Elena 291, 00161 Rome, Italy.
  • University of Roma Tre, Rome, Italy.
  • Physics Department, Harvard University, Cambridge, Massachusetts 02138, USA.
  • Cornell University, Ithaca, New York 14853, USA.

ORCID-Profile: Abhisek Ganguly, Santosh Ansumali, Sauro Succi

Förderung

  • Scuola Internazionale Superiore di Studi Avanzati
  • HORIZON EUROPE European Innovation Council
  • Abdul Kalam Technology Innovation National Fellowship

Zitieren

APA

Ganguly, A., Ansumali, S., & Succi, S. (2026). Deep neural networks as discrete dynamical systems: Implications for physics-informed learning. The Journal of Chemical Physics, 165(1). https://doi.org/10.1063/5.0315971

Vancouver

Ganguly A, Ansumali S, Succi S. Deep neural networks as discrete dynamical systems: Implications for physics-informed learning. The Journal of Chemical Physics. 2026;165(1). doi:10.1063/5.0315971

BibTeX
@article{Ganguly2026,
  author = {Ganguly, Abhisek and Ansumali, Santosh and Succi, Sauro},
  title = {Deep neural networks as discrete dynamical systems: Implications for physics-informed learning},
  journal = {The Journal of Chemical Physics},
  year = {2026},
  volume = {165},
  number = {1},
  doi = {10.1063/5.0315971}
}
Kostenlos starten →