Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MA: Fachverband Magnetismus

MA 35: PhD Focus Session: Using Artificial Intelligence Tools in Magnetism

MA 35.2: Hauptvortrag

Donnerstag, 20. März 2025, 10:05–10:35, H20

Physics meets data: decoding magnetic inhomogeneities through latent analysis — •Karin Everschor-Sitte — Faculty of Physics and Center for Nanointegration Duisburg-Essen (CENIDE), University of Duisburg-Essen

Physicists are trained to simplify complex problems into their fundamental components, often using effective minimal models to describe experimentally observed phenomena. This approach's standard challenges include the difficulty of directly measuring model parameters and the frequent oversimplification or neglect of sample inhomogeneities. As a result, models often fail to make accurate predictions when critical effects are overlooked or inadequately represented. In contrast, data-driven approaches focus on learning directly from the data, ideally without making restrictive assumptions about the data. This talk addresses the problem of hidden features in data and presents computationally efficient, physics-inspired data analysis tools - latent entropy and latent dimension [1,2] - that, for example, allow uncovering magnetic inhomogeneities from video data.

[1] I. Horenko, D. Rodrigues, T. O'Kane, K. Everschor-Sitte, Communications in Applied Mathematics and Computer Science 16, 2 (2021). [2] D. Rodrigues, K. Everschor-Sitte, S. Gerber, I. Horenko, iScience 24, 3 (2021).

Keywords: Artificial Intelligence; Latent entropy; Latent dimension; Data; Inference