SMuK 2023 – wissenschaftliches Programm

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

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 4: Neural Networks II

AKPIK 4.3: Vortrag

Mittwoch, 22. März 2023, 16:15–16:30, ZEU/0118

Partition Pooling for Convolutional Graph Network Applications in Particle Physics — •Philipp Soldin, Markus Bachlechner, Thilo Birkenfeld, Achim Stahl, and Christopher Wiebusch — III. Physikalisches Institut B, RWTH Aachen University

Convolutional Neural Networks (CNN) are often used in particle physics applications for classification and reconstruction tasks. Since the individual sensors in a particle detector are often arranged in complex geometries, the information must be projected onto regular grids to use CNNs. Convolutional Graph Networks (CGN) can encode the individual sensor positions as a static graph to prevent projection effects. However, with the number of sensors in modern particle physics detectors, the CGN performance can be limited by the considerable number of parameters. A dimensionality reduction scheme analogous to conventional pooling on images that uses graph partitioning to create pooling kernels is presented. Different CGN architectures, including partition pooling, are presented with an exemplary vertex reconstruction in an idealized neutrino detector.