Heidelberg 2022 – wissenschaftliches Programm

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

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 2: Data Analytics & Machine Learning

AKPIK 2.3: Vortrag

Mittwoch, 23. März 2022, 16:45–17:00, AKPIK-H13

Anomaly detection for Belle II PXD cluster data — •Stephanie Käs, Jens Sören Lange, Johannes Bilk, and Timo Schellhaas — Justus-Liebig-Universität Gießen

The Belle II pixeldetector (PXD) has a trigger rate of up to 30 kHz for 8 M pixels. Highly ionizing particles such as antideuterons, pions with small transverse momenta <100 MeV ("slow pions"), magnetic monopoles or stable tetraquarks generate characteristic clusters in the PXD. A large fraction of those does not reach outer detectors and therefore does not generate reconstructable tracks. We study their identification based exclusively on the PXD data, by means of anomaly detection algorithms. In this presentation, we show results from multivariate statistics analysis and tree-based multiclassifiers. In a first step, principal component analysis, linear discriminant analysis, t-distributed stochastic neighbor embedding and random forests are used for each anomaly to investigate the separability of signal and background. In a second step, a multiclassifier system shall be used. The design and output of this approach will be presented, including results on accuracy and sensitivity as well as a comparison to other methods such as Convolutional Neural Networks and Support Vector Machines.