Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 62: Trigger 1
T 62.5: Vortrag
Mittwoch, 11. März 2015, 17:45–18:00, G.10.06 (HS 6)
Studies on the Belle II L1 CDC track trigger’s z-vertex resolution with neural networks — •Sebastian Skambraks1, Sara Neuhaus1, Fernando Abudinen2, Yang Chen1, and Christian Kiesling2 — 1Technische Universität München — 2Max-Planck-Institut für Physik, München
We present the use of a neural network ensemble for the first level (L1) track trigger subsystem of Belle II. Our method employs hit and drift time information from the Central Drift Chamber (CDC). Estimating the z-coordinates of the vertex positions improves the signal to background ratio in the recorded data. Especially beam induced background can clearly be rejected, allowing to relax the 2D trigger conditions and thus enhancing the physics gain for low multiplicity events (e.g. tau pair production).
Neural networks enable an improvement of the z-vertex resolution compared to linear least squares track fitting. As general function approximators, they are capable of learning nonlinearities solely from a training dataset. We propose a combined setup, integrating the benefits of the linear fit and enriching it with the nonlinear prediction capabilities of the neural networks. The precise z-vertices of single tracks are estimated by an ensemble of local expert neural networks, specialized to sectors in the track parameter phase space. A comparison is presented, demonstrating the differences of the linear fit and the neural network approach.