Teaching
Physics-Informed Machine Learning - Materials
PhD course: Physics-Informed Machine Learning for Modeling, Planning, Control and Estimation of Physical Systems
This 5-day PhD course introduces Physics-Informed Machine Learning and its applications to modeling, planning, control, and estimation of physical systems. The course was held at the University of Trento in July 2026.
What Happens When Physics Meets Learning?
This PhD course explores how prior physical knowledge can be fused with machine learning to build models that are more interpretable, generalizable, data-efficient, and physically consistent. It bridges physics-informed learning paradigms with structured neural approaches for modeling, planning, control, and estimation, targeting autonomous and physical systems.
Topics covered:
- Physics-informed loss function paradigms
- Physics-guided input features, data, and representations
- Physics-encoded learning architectures
- Model-structured neural networks as a subclass of physics-encoded architectures
- Hands-on tutorials with the nnodely framework
- Applications to modeling, planning, control, and state estimation of physical systems
Watch the course
Watch the full course on YouTube: Physics-Informed Machine Learning playlist.
Slides
Course slides (PDF):
- Introduction
- Physics-guided inputs
- Physics-informed loss
- Physics-encoded architectures - Part I
- Physics-encoded architectures - Part II
- Physics-encoded architectures - Part III
- nnodely and applications
- Outlook and future directions
Lecturers
- Mattia Piccinini, Technical University of Munich
- Gastone Pietro Rosati Papini, University of Trento