Orateur
Description
The CLAS12 experiment is a fixed-target experiment located at the Thomas Jefferson Laboratory in Newport News, USA and dedicated to the study of hadronic physics, in particular the structure of the nucleon and hadron spectroscopy.
The Central Detector (CD) of CLAS12 provides tracking capabilities for charged-particle at large polar angles. However background hits from electronic noise or neutral particles can contaminate the clusters, distort reconstructed tracks, and lead to fake tracks. To mitigate these effects, we investigate a machine-learning approach to identify signal hits and suppress background prior to track reconstruction.
As a proof of principle, simulated events containing a single muon were used to train Graph Neural Networks (GNNs), which are particularly well suited to the irregular geometry and hit-based representation of particle detectors. Two architectures, GravNet and GarNet, have been investigated, and their performance was studied as a function of the input features and model hyperparameters.
Ongoing work focuses on validating the approach with multiple-particle events and under realistic reconstruction conditions. The impact of ML-based denoising on the performance of charged-particle tracking and its applicability to physics analyses using CLAS12 data will be shown in this talk.
| Working Group | WG5. Innovative methods for computing and reconstruction |
|---|---|
| Type of oral contribution | Experiment |