Cargando…

Adversarial training for b-tagging algorithms in CMS

Modern neural networks bring considerable performance improvements in various areas of high-energy physics, such as object identification. Flavour-tagging is one example that profits from complex architectures, leveraging information from large numbers of low-level inputs. While such tagging algorit...

Descripción completa

Detalles Bibliográficos
Autor principal: CMS Collaboration
Lenguaje:eng
Publicado: 2022
Materias:
Acceso en línea:http://cds.cern.ch/record/2839919
_version_ 1780975997540106240
author CMS Collaboration
author_facet CMS Collaboration
author_sort CMS Collaboration
collection CERN
description Modern neural networks bring considerable performance improvements in various areas of high-energy physics, such as object identification. Flavour-tagging is one example that profits from complex architectures, leveraging information from large numbers of low-level inputs. While such tagging algorithms are evaluated on data and simulation for analysis purposes, training is usually performed with simulated samples only. Differences in performance between these two domains are observed which need to be calibrated against. With a new strategy, called adversarial training, we reduce the observed differences prior to any calibration, and improve robustness of the classifier against injected mis-modelings that mimic systematic uncertainties. In this note, studies on adversarial robustness and agreement between data and simulation are carried out with the DeepJet algorithm, evaluated on Run2 samples. The addition of an adversarial module is envisaged to be included in newly developed tagging algorithms for Run3.
id cern-2839919
institution Organización Europea para la Investigación Nuclear
language eng
publishDate 2022
record_format invenio
spelling cern-28399192022-11-08T22:14:33Zhttp://cds.cern.ch/record/2839919engCMS CollaborationAdversarial training for b-tagging algorithms in CMSDetectors and Experimental TechniquesModern neural networks bring considerable performance improvements in various areas of high-energy physics, such as object identification. Flavour-tagging is one example that profits from complex architectures, leveraging information from large numbers of low-level inputs. While such tagging algorithms are evaluated on data and simulation for analysis purposes, training is usually performed with simulated samples only. Differences in performance between these two domains are observed which need to be calibrated against. With a new strategy, called adversarial training, we reduce the observed differences prior to any calibration, and improve robustness of the classifier against injected mis-modelings that mimic systematic uncertainties. In this note, studies on adversarial robustness and agreement between data and simulation are carried out with the DeepJet algorithm, evaluated on Run2 samples. The addition of an adversarial module is envisaged to be included in newly developed tagging algorithms for Run3.CMS-DP-2022-049CERN-CMS-DP-2022-049oai:cds.cern.ch:28399192022-10-24
spellingShingle Detectors and Experimental Techniques
CMS Collaboration
Adversarial training for b-tagging algorithms in CMS
title Adversarial training for b-tagging algorithms in CMS
title_full Adversarial training for b-tagging algorithms in CMS
title_fullStr Adversarial training for b-tagging algorithms in CMS
title_full_unstemmed Adversarial training for b-tagging algorithms in CMS
title_short Adversarial training for b-tagging algorithms in CMS
title_sort adversarial training for b-tagging algorithms in cms
topic Detectors and Experimental Techniques
url http://cds.cern.ch/record/2839919
work_keys_str_mv AT cmscollaboration adversarialtrainingforbtaggingalgorithmsincms