SMuK 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 9: AI Topical Day – New Methods (joint session AKPIK/T)
AKPIK 9.4: Vortrag
Donnerstag, 23. März 2023, 18:15–18:30, HSZ/0004
Photon identification at hadron colliders using graph neural networks — •Ali Malyali Choban1, Johannes Erdmann1, Florian Mausolf1, and Christopher Morris2 — 1III. Physikalisches Institut A, RWTH Aachen University — 2Fachgruppe Informatik, RWTH Aachen University
At hadron colliders like the LHC, photons are essential physics objects in a wide range of analyses. For example, they allow the study of the Higgs boson using the diphoton decay channel. At a typical particle detector, the main signatures of photons are energy depositions in the electromagnetic calorimeter. However, other objects can leave similar signatures in the electromagnetic calorimeter, leading to misidentification as photons. Jets are abundant at the LHC and they include a high number of light hadrons, most notably neutral pions decaying into two photons. The decay of pions produces photons that are often close to each other and they are likely to be reconstructed as a single photon. However, photon candidates from jets have different attributes that can help to discriminate them from real photons. Specifically, they tend to produce wider signatures in the calorimeter, and to be accompanied by more additional particles.
Graph neural networks (GNNs) are flexible neural architectures well suited for dealing with input data of irregular structure and variable shape. Hence, they are particularly suited for classifying photon candidates as often a variable number of particles surrounds them. In this talk, our study of the applicability of GNNs for photon identification and comparisons with convolutional neural networks are presented.