SMuK 2021 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

HK: Fachverband Physik der Hadronen und Kerne

HK 14: Hadron Structure and Spectroscopy III

HK 14.4: Talk

Tuesday, August 31, 2021, 17:30–17:45, H3

Improving Kaon-Pion Identification with Machine Learning Techniques for CLAS12 — •Áron Kripkó, Stefan Diehl, and Kai-Thomas Brinkmann for the CLAS collaboration — II. Physikalisches Institut, Justus Liebig Universität Gießen, 35392 Gießen, Germany

For semi-inclusive deep inelastic scattering (SIDIS), a reliable particle identification and background estimation is a key requirement. This is especially true for Kaon SIDIS, where a strong pion contamination can be expected at high momenta if only time of flight measurements are used for the particle identification. For the SIDIS Kaon production from the scattering of 10.6 GeV electrons in the recently upgraded CLAS12 detector, Kaons are hardly distinguishable from pions above 3 GeV with a pure time of flight based PID. Currently a RICH detector, which will provide good separation above 3 GeV is only available in one sector. Here advanced PID methods based on neural networks, exploiting the information from all detector components can help improve the situation.

In this talk machine learning methods, which could complement the other traditional particle identification methods, are described and compared. Based on multiple independent checks, the best method can efficiently reduce the pion contamination in the kaon sample in the whole momentum range, by still keeping the statistics on a reasonable level.

This work is supported by HFHF.