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.
Highlights
Modern machine learning has recently gained significant impact in many directions of LHC physics
This is a natural step given our improved understanding of subjet physics both experimentally and theoretically, combined with the rapid development of standard machine learning tools
We study the ability of the network to track uncertainties due to the limited size of the training sample and due to systematics like the jet energy scale
Summary
Modern machine learning has recently gained significant impact in many directions of LHC physics. Multi-variate analyses of high-level observables are currently being replaced by deep neural networks with low-level observables This is a natural step given our improved understanding of subjet physics both experimentally and theoretically, combined with the rapid development of standard machine learning tools. The remaining open questions, which need to be studied before we can widely apply these kinds of taggers to standard LHC analyses, are related to systematics [25], general uncertainties [26], stability, weakly supervised learning [27,28,29,30,31,32,33,34], understanding the relevant physics input [35,36,37,38,39], and other LHC-specific issues which do not automatically have a counterpart in modern machine learning. We show how the Bayesian network offers a new handle to test the stability of a classification network, for instance in the presence of pile-up
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
