Astronomy16 papers
Artificial Intelligence and Machine Learning in Planetary Science
A companion collection for the ACM 2026 talk by Valerio Carruba
VC
Valerio Carruba
All Reviews
AstronomyMust Read
intermediate
Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
Raissi, M. et al. (2019)
- Published
- Feb 1, 2019
- Journal
- Journal of Computational Physics · Vol. 378
- DOI
- 10.1016/j.jcp.2018.10.045
At a GlanceAI
Physics-informed neural networks use PDE constraints to solve and infer nonlinear dynamics from data in one framework.
SummaryAI
This paper introduced physics-informed neural networks (PINNs), a way to train neural networks using both data and the governing equations of a physical system. The key advance is that the model embeds nonlinear partial differential equation constraints directly into learning, allowing it to address both forward prediction and inverse parameter identification. This helped bridge machine learning and scientific computing, with implications for solving complex physical problems when data are limited or noisy.
Method SnapshotAI
The approach uses neural networks trained with loss terms that enforce the governing nonlinear partial differential equations alongside data fitting.
BackgroundAI
Background in differential equations, scientific computing, and basic neural networks is helpful.
A landmark paper that introduced Physics-Informed Neural Networks (PINNs), establishing one of the most influential frameworks for integrating physical laws with deep learning. By embedding differential equation constraints directly into neural network training, this work opened new avenues for data-driven scientific computing and inverse modeling. Essential reading for researchers interested in applying machine learning to physics, engineering, and celestial mechanics.
— VC