München 2019 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
HK: Fachverband Physik der Hadronen und Kerne
HK 44: Instrumentation IX
HK 44.7: Talk
Wednesday, March 20, 2019, 18:15–18:30, HS 11
Particle-Antiparticle Discrimination Using Neural Networks — Laura Fabbietti, Martin J. Losekamm, •Jan Henrik Müller, Stephan Paul, and Thomas Pöschl — Technische Universität München
Cosmic-ray antiproton measurements are challenging because of the small flux of antiprotons in comparison to the large background flux of ordinary ions, requiring an effective particle-identification algorithm. A classical approach is to fully reconstruct the event topology to draw conclusions about the incoming particle’s species. However, for space-based experiments, this can be impracticable since the available computing power and calculation time are not sufficient to allow such complex calculations. An alternative approach is the use of neural networks: They can be trained on high-power computers on ground and then implemented on the detector’s front-end electronics to provide fast reconstruction with small computational effort. In this work, we assess different types of neural networks for identifying antiprotons, protons, and heavy ions in the Multi-Purpose Active-Target Particle Telescope and evaluate their performance with simulated detector data. We analyze the result, focusing on events that were wrongly classified. This is necessary because the signal-to-background ratio of antiprotons is in the order of 10−4 and rare misclassifications have a large impact on the flux measurement. We also discuss possibilities to increase the network’s reconstruction ability for rare events.