Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 8: Machine Learning in Dynamics and Statistical Physics II (joint session DY/SOE)
SOE 8.7: Vortrag
Dienstag, 19. März 2024, 11:00–11:15, BH-N 243
Data assimilation of cardiac dynamics by means of adjoint optimization — •Inga Kottlarz1,2,3,4, Sebastian Herzog2,4,5, Patrick Vogt2,3, Stefan Luther1,2,4, and Ulrich Parlitz2,3,4 — 1Institute for Pharmacology and Toxicology, UMG Göttingen, Germany — 2MPI for Dynamics and Self-Organization, Göttingen, Germany — 3Institute for the Dynamics of Complex Systems, University of Göttingen, Germany — 4German Center for Cardiovascular Research, Partner Site Niedersachsen, Göttingen, Germany — 5III. Institute of Physics, University of Göttingen, Germany
Cardiac muscle tissue is an excitable medium that can exhibit a range of dynamics of different complexity, from planar waves to spiral waves to spatiotemporal chaos, the latter being associated with (fatal) cardiac arrhythmia.
Both the prediction of such high dimensional chaotic time series, as well as the reconstruction of their (not yet fully observable) complete dynamical state are ongoing challenges. In recent years, machine learning approaches have gained popularity for solving these problems, which can be advantageous if we do not have much knowledge about the dynamical system in question, but are limited by the large amounts of training data that is needed and often not available for biological systems. We present adoptODE, an adjoint optimization framework for estimating model parameters and unobserved variables. We showcase the adjoint method*s effectiveness in optimizing high-dimensional problems with thousands of unknowns, serving as a valuable tool for bridging the gap between empirical data and theoretical models.
Keywords: adjoint optimization; physics informed machine learning; cardiac dynamics; data assimilation