Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 34: Poster: Machine Learning, Data Science, and Reservoir Computing
DY 34.6: Poster
Mittwoch, 20. März 2024, 15:00–18:00, Poster C
Sand Grain Generation through Deep Learning and Lower Dimensional Representations — •Lira Yelemessova1 and Matthias Schröter1,2 — 1Georg-August-Universität Göttingen, Göttingen, Germany — 2Max Planck Institute for Dynamics and Self-Organization, Göttingen, Germany
How would one create artificial realistic sand grains? This project explores this question by representing sand grains as point clouds and then employing the denoising diffusion probabilistic model. The first step is to use an autoencoder to transform the complex three-dimensional structures of synthetic sand grains into a lower-dimensional space. Then, the model generates additional samples using denoising diffusion, which is also the algorithm behind programs such as Stable Diffusion and DALL-E. We study how variations in the number of points and dimensions of additional features impact the generated samples.
Keywords: granular matter; neural network; autoencoder; shape