Heidelberg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 25: Data Analysis, Information Technology and Artificial Intelligence
T 25.4: Vortrag
Montag, 21. März 2022, 17:10–17:25, T-H38
Evaluating Uncertainties in Measurements of the Production of a Single Top-Quark in Association with a Photon with Bayesian Neural Networks — Johannes Erdmann1, Burim Ramosaj2, and •Daniel Wall1 — 1TU Dortmund University, Department of Physics — 2TU Dortmund University, Department of Statistics
Multivariate approaches including neural networks constitute powerful and established methods in experimental particle physics. However, using these methods, it is difficult to account for uncertainties from statistical and systematic sources in a consistent and efficient way. By employing weight distributions instead of fixed weights and by utilising the process of Bayesian inference, Bayesian Neural Networks not only suffer significantly less from overfitting, but also allow to obtain an uncertainty estimate on the output.
These characteristics are of particular interest in measurements of processes suffering from limited statistics and challenging signal-to-background ratios. The analysis of top-quark production in association with a photon (tqγ), probing the structure of the electroweak couplings of the top quark, is one of such processes, as the corresponding cross section is considerably lower than those of relevant background processes, most importantly top-quark pair production (ttγ).
In this talk, studies of Bayesian Neural Networks for their application in the classification of top-quark processes in association with a photon are presented.