Deep-learning jets with uncertainties and more

• Verfasst von: Bollweg, Sven [VerfasserIn]
• Verfasst von: Haußmann, Manuel [VerfasserIn]
• Verfasst von: Kasieczka, Gregor [VerfasserIn]
• Verfasst von: Luchmann, Michel [VerfasserIn]
• Verfasst von: Plehn, Tilman [VerfasserIn]
• Verfasst von: Thompson, Jennifer M. [VerfasserIn]
• Titel: Deep-learning jets with uncertainties and more
• Verf.angabe: Sven Bollweg, Manuel Haussmann, Gregor Kasieczka, Michel Luchmann, Tilman Plehn and Jennifer Thompson
• E-Jahr: 2020
• Jahr: 16-01-2020
• Umfang: 25 S.
• Fussnoten: Gesehen am 27.02.2020
• Titel Quelle: Enthalten in: SciPost physics
• Ort Quelle: Amsterdam : SciPost Foundation, 2016
• Jahr Quelle: 2020
• Band/Heft Quelle: 8(2020), 1, Artikel-ID 006, Seite 1-25
• ISSN Quelle: 2542-4653
• DOI: doi:10.21468/SciPostPhys.8.1.006
• Datenträger: Online-Ressource
• Sprache: eng
• K10plus-PPN: 1691153532

Abstract

Bayesian neural networks allow us to keep track of uncertainties, for example in top tagging, by learning a tagger output together with an error band. We illustrate the main features of Bayesian versions of established deep-learning taggers. We show how they capture statistical uncertainties from finite training samples, systematics related to the jet energy scale, and stability issues through pile-up. Altogether, Bayesian networks offer many new handles to understand and control deep learning at the LHC without introducing a visible prior effect and without compromising the network performance.