Scientific talk by Alexandre Martin
Abstract: Geometric Deep Learning (GDL) represents a major advance in the analysis of geometrically structured data, by hardcoding the principles of symmetry and invariance into architectures. Unlike traditional deep learning architectures, which are mainly adapted to Euclidean data, GDL ones extend to non-Euclidean domains such as graphs, varieties and point clouds. My presentation will explore the theoretical foundations of GDL, highlighting the importance of symmetries – translation, rotation, permutation – in designing architectures that are both efficient and robust.
I will illustrate these concepts through an application to the classification of organoids, three-dimensional multicellular structures used in biomedical research. By modelling organoids in the form of graphs, where the nodes represent the cells and the edges reflect their spatial or functional relationships, it is possible to use GDL tools to identify phenotypic signatures.
When: Monday, December 16 at 2pm
Where: Euler Bleu
