SMuK 2023 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 72: Exp. Methods II

T 72.3: Vortrag

Mittwoch, 22. März 2023, 16:20–16:35, POT/0106

Graph building and input feature analysis for edge classification in the Central Drift Chamber at Belle II — •Philipp Dorwarth, Torben Ferber, Lea Reuter, and Slavomira Stefkova — Institute of Experimental Particle Physics (ETP), Karlsruhe Institute of Technology (KIT)

Many extensions of the Standard Model, such as inelastic dark matter models, predict long-lived particles. They can manifest with two charged tracks originating from a vertex with a large displacement from the interaction point in collider experiments. Conventional tracking algorithms are insufficient to respond to those highly displaced vertices, and they also scale poorly with an increased beam background, as expected from SuperKEKB’s increased luminosity.

Graphs are an intuitive representation of hits in a tracking detector as they provide high flexibility regarding input features and the length of input vectors. Therefore, we develop a Graph Neural Network (GNN) approach for hit and edge classification in the Central Drift Chamber (CDC) at Belle II. Eventually, the output will be used for GNN-based displaced vertex and tracking algorithms. We examine different methods of graph building and analyze their performance for the classification task. In addition, we study the feasibility of using detector-level information, such as digitized signal hits, as GNN input features in both data and simulation. We find that this information provides very good discriminatory power and should therefore be used as an additional input feature for the GNN to improve the efficiency of the edge classification.