Die DPG-Frühjahrstagung in Bonn musste abgesagt werden! Lesen Sie mehr ...

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

HK: Fachverband Physik der Hadronen und Kerne

HK 20: Instrumentation III

HK 20.4: Vortrag

Dienstag, 31. März 2020, 17:45–18:00, J-HS C

Machine Learning Approach for Track Finding Using Language Models — •Jakapat Kannika, James Ritman, and Tobias Stockmanns — Forschungszentrum Jülich, Jülich, Germany

In the particle physics experiments, track finding is a pattern recognition task in which input hits are clustered into different groups of output tracks. The hits are signals of the particles traveling through the detectors, and the tracks are groups of trajectories of those particles. This study is focusing on implementing a track finding algorithm using language models for straw tube based tracking systems. The language model is a probability distribution which is used in order to recognize the sequences of data. The model is widely used in the field of natural language processing, where applications such as speech recognition, handwriting recognition, word prediction also use the language models. In the current study, we extract features from the hit data and treat them as discrete values similarly to words, then do a language modeling. The obtained language model is used in the same way as in the word prediction applications, but in this case, it predicts the next hits. The algorithm is now able to learn how to distinguish between true hit and noise, and it can also recognize tracks with long dependency patterns. The current status and an outlook on the overall performance will be presented.