IPAB Workshop - 23/4/26 Title: Theory-trained neural networks (a.k.a. PINNs)Abstract: Theoretical descriptions often don't match the complexity of practical problems, and real-world data tend to provide only partial views on the general situation. Addressing this dilemma, a class of networks trained by theory (TTNs, initially as physics-informed neural networks or PINNs) has become available in the last decade, providing methods for machine learning within the range between small data and big data. The talk will highlight several examples and application cases to demonstrate that TTNs are useful even for inexact knowledge and sparse data. In a second part, we consider explainability and interpretability of TTNs based on their information theory and training paradigms and consider combinations of TTNs with other machine learning methods. Apr 23 2026 13.00 - 14.00 IPAB Workshop - 23/4/26 Michael Herrmann MF2
IPAB Workshop - 23/4/26 Title: Theory-trained neural networks (a.k.a. PINNs)Abstract: Theoretical descriptions often don't match the complexity of practical problems, and real-world data tend to provide only partial views on the general situation. Addressing this dilemma, a class of networks trained by theory (TTNs, initially as physics-informed neural networks or PINNs) has become available in the last decade, providing methods for machine learning within the range between small data and big data. The talk will highlight several examples and application cases to demonstrate that TTNs are useful even for inexact knowledge and sparse data. In a second part, we consider explainability and interpretability of TTNs based on their information theory and training paradigms and consider combinations of TTNs with other machine learning methods. Apr 23 2026 13.00 - 14.00 IPAB Workshop - 23/4/26 Michael Herrmann MF2
