Physics-Informed Machine Learning and Hybrid Modelling

Organizers

Bernhard Geiger*,+, Manfred Mücke#, Stefan Posch^

* Signal Processing and Speech Communication Laboratory, Graz University of Technology
+ Know Center Research GmbH
# Materials Center Leoben Forschung GmbH
^ Christian Doppler Laboratory for Physics-driven Machine Learning in Industrial Applications, Graz University of Technology

Workshop Description

The objective of this workshop is to present, explore, and critically discuss recent advancements in the rapidly evolving field of physics-based machine learning (PIML) and hybrid modeling. This interdisciplinary domain merges traditional physics-driven numerical methods with modern machine learning techniques, aiming to improve model fidelity, reduce computational cost, and enhance generalizability across a wide range of scientific and engineering problems such as fluid dynamics, solid mechanics, communications, or computational medicine.

Hybrid modeling, in particular, leverages the strengths of both paradigms—combining first-principles models with machine learning—to overcome limitations inherent in purely data-driven or purely mechanistic approaches. A fundamental challenge in hybrid modeling is to understand the propagation of errors and uncertainties of the (data-driven or first principles) parts, and how this affects the qualitative and quantitative behavior of the hybrid model. Complementary to hybrid modeling, PIML utilizes first-principles knowledge in the creation of machine learning models, influencing data selection, model parameterization, or learning itself via regularization. The workshop shall serve as a platform to connect developments in fundamental theory, algorithmic innovation, and application-driven research.

We invite contributions addressing the following topics:

  • Hybrid modeling approaches combining laws of physics and machine learning
  • Neural-enhanced algorithms and model-based deep learning
  • Novel physics-informed training objectives
  • Model parameterization approaches based on prior knowledge
  • Methods for constructing probabilistic hybrid or PIML models
  • Domain-aware training data creation/selection
  • Applications of PIML and hybrid models in science and engineering
  • Uncertainty and error propagation in hybrid models
  • Fundamental trade-offs in PIML and hybrid modeling (energy-performance, complexity-accuracy, etc.)

In addition to original research, we also explicitly invite extended abstracts summarizing already published results or work in progress.

An important aim of this workshop is to connect researchers in the field of PIML and hybrid modeling, and to thus establish a strong community in this field. Being the first workshop of its kind at AIRoV, we aim to identify the needs and common research interests of this community and to develop plans for joint collaborations during a discussion session.

Tentative Workshop Schedule

  • Welcome by the organizers, introductory lecture, and workshop outline: 30 mins
  • Keynote: 30 mins
  • Contributed Talks 1-2: 20 mins each
  • Coffee Break: 20 mins
  • Contributed Talks 3-5: 20 mins each
  • Spotlight Talks 1-8: 5 mins each
  • Coffee Break: 20 mins
  • Discussion Session: 120 mins

Keynote: Nils Thuerey -- Differentiable PDE Solvers for Numerical Simulations

The talk will present lessons from the area of AI/deep learning for physics simulations with numerical solvers. A key focus is the utilization of solvers that are capable of providing Jacobians, i.e. "differentiable simulations". These solvers seamlessly integrate with deep learning algorithms, presenting numerous advantages in arising from AI-based components in solvers, particularly in the context of flow simulations. However, the availability of gradient computation is not ubiquitous in many existing simulation environments. Avenues for computing them will be discussed, as well as fallbacks if they're not available (yet). It will be shown how to leverage differentiable solvers in varied applications such as closure modeling for Navier-Stokes, accelerated solving and inverse problems.

Prof. Thuerey works in the field of computer graphics, with a particular emphasis on physics-based deep learning algorithm. One focus of his research targets the simulation of fluid phenomena, such as water and smoke. These simulations find applications as visual effects in computer generated worlds, but also in many fields of engineering. Examples of his work are novel algorithms to make simulations easier to control, to handle detailed surface tension effects, and to increase the amount of turbulent detail. After studying computer science, Prof. Thuerey acquired a PhD for his work on liquid simulations in 2006. He received both degrees from the University of Erlangen-Nuremberg. Until 2010 he held a position as a post-doctoral researcher at ETH Zurich, in collaboration with Ageia/Nvidia. Subsequently, he worked for three years as Research & Development Lead at ScanlineVFX, developing large scale physics-simulators for visual effects. Since fall 2013 he has been Professor for Physics-based Simulation at TUM.

Paper Submission

Please refer to the Call for Papers for details regarding the double blind submission.

Important Dates

  • Paper submission deadline: 2026-03-14
  • Notification of acceptance: 2026-03-24

Contact

E-Mail: geiger@tugraz.at