Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 33: Machine Learning in Dynamics and Statistical Physics I

DY 33.10: Talk

Thursday, March 20, 2025, 12:00–12:15, H47

Data-Driven Sparse Identification with Adaptive Function Bases — •Gianmarco Ducci, Maryke Kouyate, Karsten Reuter, and Christoph Scheurer — Fritz-Haber-Institut der MPG, Berlin

Interpretable data-driven methods have proven viable for deriving kinetic equations directly from experimental data. However, such numerical methods are inherently susceptible to noise, which affects the sparsity in the resulting models. In order to promote such a sparsity condition, finding the optimal set of basis functions is a necessary prerequisite, but yet a challenging task to determine in advance.

We here present our in-house developed ddmo (Data-Driven Model Optimizer) software, which allows precise control over the space of candidate constituent terms. Such a complete framework comprises two main novel features. The first feature permits to include parametric functions in the library. The second feature is an adaptive library sizing routine that progressively adds or removes elements based on the learning from the dataset. We show a practical application of our algorithm tailored at identifying Langmuir-Hinshelwood mechanisms from experimental data.

Keywords: Data-Driven Modelling; Sparse Identification