SKM 2023 – wissenschaftliches Programm

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

DY: Fachverband Dynamik und Statistische Physik

DY 17: Machine Learning in Dynamics and Statistical Physics I

DY 17.6: Vortrag

Dienstag, 28. März 2023, 11:30–11:45, ZEU 160

Classification of Gel Networks using Graph Convolutional Neural Networks — •Matthias Gimperlein and Michael Schmiedeberg — FAU Erlangen-Nürnberg, Erlangen, Germany

The structural properties of gel networks are important for the mechanical properties of the corresponding gels. We analyze gel networks and their structure using a machine learning approach based on graph convolutional networks (GCN) employing only the local neighborhood of particles as input information.

Using these we define a GCN-Autoencoder to reconstruct adjacency matrices of networks and quantitatively analyze in which properties the prediction of the network differs from the original input. This includes analysis on the abstract graph level as well as on the real physical network level.

Furthermore we use GCNs to classify gel networks depending i.e. on the loopsizes which are present in the network. Our goals include getting robust classification of strongly or weakly connected gel networks, predictions of minimal connecting structures and an insight how - according to an artificial intelligence - gel networks look like.