The simulation of complex systems as well as the reconstruction of the state of that system from experimental data is a very time consuming task. Workshop leader Dr. Nico Hoffmann from HZDR and TUD together with 13 participants from different institutions will be approaching this challenge by Physics-informed neural network allowing us to approximate forward simulations as well as identify the governing equations and parameters based on experimental measurements.

The workshop is driven by the following question: are we able to either efficiently solve or recover the governing equations of certain complex systems in terms of Ordinary or Partial Differential Equations (ODE/PDE) by Physics-informed neural networks?

PINN basically provide all means for jointly solving the following tasks

• solving 1D/2D problems provided initial/boundary conditions + governing equations (PINN can be seen as implicit solver of ODE/PDEs by optimization),
• warm starting simulations by transfer learning (i.e. recovering the solution of a changed configuration),
• learning the potentially non-linear contribution of certain parameters to the solution of a complex model,
• identification of parameters and/or governing equations provided either full simulation- or experimental data.
• The workshop consists of theory as well as hackathon sessions. The theory sessions introduce basic and advanced concepts of Physics-informed neural networks while the Hackathon sessions enable the participants to adapt the PINN-framework of HZDR’s Hoffmann lab to very specific research questions ranging from quantum mechanics to medical imaging and materials science.