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

T: Fachverband Teilchenphysik

T 87: Datenanalyse

T 87.1: Vortrag

Donnerstag, 22. März 2018, 16:30–16:45, Z6 - SR 2.005

Adversarial networks used in a single-top-quark analysis in ATLAS — •Rui Zhang and Ian C. Brock — Rheinische Friedrich-Wilhelms-Universität Bonn

Multivariate analysis (MVA) techniques are widely used in high energy physics to separate interesting signal processes from a large amount of background events. The training of the MVA is usually done using nominal signal and background samples. However, the imperfect knowledge of the detector performance and physics model results in the presence of systematic uncertainties that affect the classifier. A step further would be to construct a classifier insensitive to systematic variations, which are usually parametrised by nuisance parameters (NP). Adversarial networks, which consist of a system of neural networks contesting each other, are a clear candidate to solve such problem. This talk will investigate the possibilities of using this technique in a single-top-quark analysis in ATLAS. Monte Carlo events are split into training and test samples for both nominal and systematic variations. Adversarial networks are built by Keras, where the discriminative network is trying to distinguish signal and backgrounds while cheating the generative network, which tries to predict NP values.