Hauptinhalt
Topinformationen
Geometric Deep Learning
DozentIn: Prof. Dr. Marcel Campen
Veranstaltungstyp: Vorlesung und Übung
Ort: 32/109
Zeiten: Di. 16:00 - 18:00 (wöchentlich) - Vorlesung, Mi. 14:00 - 16:00 (wöchentlich) - Vorlesung/Übung
Beschreibung: Fundamentals of geometric aspects in machine learning (e.g. invariance, equivariance, multi-scale structure). Overview over representational options (e.g. point clouds, grids, meshes, implicits, parametrics) and corresponding challenges together with advanced technical approaches to learning over and learning of geometric data, in particular 3D objects and scenes. Basic concepts involved in this context include artificial neural networks, convolution, pooling, diffusion, continuous convolution, random walks, transformers, generative approaches. Case studies include shape classification, shape segmentation, shape correspondence, shape generation.