Münster 2017 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 106: Suche nach Dunkler Materie 4

T 106.6: Vortrag

Donnerstag, 30. März 2017, 18:10–18:25, VSH 19

Artificial Neural Networks as event classifiers for an InGrid detector at CAST — Klaus Desch, Jochen Kaminski, Christoph Krieger, Tobias Schiffer, and •Sebastian Schmidt — Physikalisches Institut der Universität Bonn, Deutschland

The CERN Axion Solar Telescope is a helioscope experiment at CERN searching for solar axions and chameleons. In the magnetic field of a decommissioned LHC prototype dipole magnet the particles are reconverted to photons via the inverse Primakoff effect. The resulting photons are in the low X-ray regime. A low conversion probability means the data is dominated by background events. Thus, methods to differentiate between background and real X-ray events need to be very efficient, providing a very high background suppression, but still retain high signal efficiency.

In 2014 and 2015 a gaseous detector based on a single InGrid was deployed. A likelihood method was used in our analysis of this data. In this talk an approach using Convolutional Neural Networks (CNNs) will be presented. These are widely used in commercial applications, e.g. for image classification. CNNs allow to use the raw individual frames as inputs, without any need for preprocessing.

The talk will briefly cover the basics of Artificial Neural Networks (ANNs), compare the usage of standard ANNs in particle physics with a CNN, explain the implementation and present some preliminary results.