V-tagger with jet mass de-corellation
A machine-learning based V-boson tagger was developed in the last few years. The tagger is trained on a set of UFO large-radius (R=1.0) jets substructure variables which are known to have a discriminant power in distinguishing jets originated from a heavy 2-prong resonance from jets originated from the usual QCD background (low-mass quark/gluon). A recent PUBnote with further details was produced ATL-PHYS-PUB-2021-029 . We refer to such PUBnote for further details. Let us remind that the tagger is available for UFO jets with: With respect to the published note, the network was re-trained to correct for a couple of minor issues discovered in very first analysis checks:- bug in the calculation of KtDr jet substructure variable
- mass de-correlation at low jet pT
- for the mass de-corellated tagger, it was observed the presence of a residual mass shaping in the tagged QCD spectrum. This is an unwanted behavior for analyses relying on bump-hunting on smooth background mass distributions.
Implementation in athena
The taggers are available to be used in athena through the BoostedJetTaggers package. Input and configuration files for the BoostedJetTaggers are available at /afs/cern.ch/work/d/dmelini/public/Wtagger_KtDrFix/KtDr fix
The KtDr variable is one of the variables used in the training if the ML-based V-tagger.The distribution of KtDr (phi) before and after the fix in the KtDr calcualtion, for jets used in the training of the network, can be seen in the plot below. The calculation of KtDr was fixed in the tool computing the jet variables and is not a problem anymore if you are using a recent Athena version.
Mass correlation at low pT
It was tried to reduce the shaping observed at low pT, unfortunately with not much success. On the other way, it was investigated in more detail and it was found that much of the tagger working points were not strongly affected and in general for jet pT > 300 GeV the mass shaping is much reduced. We think this piece of information could be of interest for analyses planning to use this tagger.| 200 GeV < pT < 500 GeV | 230 GeV < pT < 280 GeV | 250 GeV < pT < 300 GeV | 300 GeV < pT < 500 GeV | 500 GeV < pT < 1000 GeV | 1000 GeV < pT < 2000 GeV | |
|---|---|---|---|---|---|---|
| 50% sig.eff. WP | __pT200_500_unnorm.pdf.png "jetmasscomparison__eff_sig_50__myNN_myANN(lambda=10)__pT200_500_unnorm.pdf.png") |
__pT230_280_unnorm.pdf.png "jetmasscomparison__eff_sig_50__myNN_myANN(lambda=10)__pT230_280_unnorm.pdf.png") |
__pT250_300_unnorm.pdf.png "jetmasscomparison__eff_sig_50__myNN_myANN(lambda=10)__pT250_300_unnorm.pdf.png") |
__pT300_500_unnorm.pdf.png "jetmasscomparison__eff_sig_50__myNN_myANN(lambda=10)__pT300_500_unnorm.pdf.png") |
__pT500_1000_unnorm.pdf.png "jetmasscomparison__eff_sig_50__myNN_myANN(lambda=10)__pT500_1000_unnorm.pdf.png") |
__pT1000_2000_unnorm.pdf.png "jetmasscomparison__eff_sig_50__myNN_myANN(lambda=10)__pT1000_2000_unnorm.pdf.png") |
| 60% sig.eff. WP | __pT200_500_unnorm.pdf.png "jetmasscomparison__eff_sig_60__myNN_myANN(lambda=10)__pT200_500_unnorm.pdf.png") |
__pT230_280_unnorm.pdf.png "jetmasscomparison__eff_sig_60__myNN_myANN(lambda=10)__pT230_280_unnorm.pdf.png") |
__pT250_300_unnorm.pdf.png "jetmasscomparison__eff_sig_60__myNN_myANN(lambda=10)__pT250_300_unnorm.pdf.png") |
__pT300_500_unnorm.pdf.png "jetmasscomparison__eff_sig_60__myNN_myANN(lambda=10)__pT300_500_unnorm.pdf.png") |
__pT500_1000_unnorm.pdf.png "jetmasscomparison__eff_sig_60__myNN_myANN(lambda=10)__pT500_1000_unnorm.pdf.png") |
__pT1000_2000_unnorm.pdf.png "jetmasscomparison__eff_sig_60__myNN_myANN(lambda=10)__pT1000_2000_unnorm.pdf.png") |
| 70% sig.eff. WP | __pT200_500_unnorm.pdf.png "jetmasscomparison__eff_sig_70__myNN_myANN(lambda=10)__pT200_500_unnorm.pdf.png") |
__pT230_280_unnorm.pdf.png "jetmasscomparison__eff_sig_70__myNN_myANN(lambda=10)__pT230_280_unnorm.pdf.png") |
__pT250_300_unnorm.pdf.png "jetmasscomparison__eff_sig_70__myNN_myANN(lambda=10)__pT250_300_unnorm.pdf.png") |
__pT300_500_unnorm.pdf.png "jetmasscomparison__eff_sig_70__myNN_myANN(lambda=10)__pT300_500_unnorm.pdf.png") |
__pT500_1000_unnorm.pdf.png "jetmasscomparison__eff_sig_70__myNN_myANN(lambda=10)__pT500_1000_unnorm.pdf.png") |
__pT1000_2000_unnorm.pdf.png "jetmasscomparison__eff_sig_70__myNN_myANN(lambda=10)__pT1000_2000_unnorm.pdf.png") |
| 80% sig.eff. WP | __pT200_500_unnorm.pdf.png "jetmasscomparison__eff_sig_80__myNN_myANN(lambda=10)__pT200_500_unnorm.pdf.png") |
__pT230_280_unnorm.pdf.png "jetmasscomparison__eff_sig_80__myNN_myANN(lambda=10)__pT230_280_unnorm.pdf.png") |
__pT250_300_unnorm.pdf.png "jetmasscomparison__eff_sig_80__myNN_myANN(lambda=10)__pT250_300_unnorm.pdf.png") |
__pT300_500_unnorm.pdf.png "jetmasscomparison__eff_sig_80__myNN_myANN(lambda=10)__pT300_500_unnorm.pdf.png") |
__pT500_1000_unnorm.pdf.png "jetmasscomparison__eff_sig_80__myNN_myANN(lambda=10)__pT500_1000_unnorm.pdf.png") |
__pT1000_2000_unnorm.pdf.png "jetmasscomparison__eff_sig_80__myNN_myANN(lambda=10)__pT1000_2000_unnorm.pdf.png") |
| 90% sig.eff. WP | __pT200_500_unnorm.pdf.png "jetmasscomparison__eff_sig_90__myNN_myANN(lambda=10)__pT200_500_unnorm.pdf.png") |
__pT230_280_unnorm.pdf.png "jetmasscomparison__eff_sig_90__myNN_myANN(lambda=10)__pT230_280_unnorm.pdf.png") |
__pT250_300_unnorm.pdf.png "jetmasscomparison__eff_sig_90__myNN_myANN(lambda=10)__pT250_300_unnorm.pdf.png") |
__pT300_500_unnorm.pdf.png "jetmasscomparison__eff_sig_90__myNN_myANN(lambda=10)__pT300_500_unnorm.pdf.png") |
__pT500_1000_unnorm.pdf.png "jetmasscomparison__eff_sig_90__myNN_myANN(lambda=10)__pT500_1000_unnorm.pdf.png") |
__pT1000_2000_unnorm.pdf.png "jetmasscomparison__eff_sig_90__myNN_myANN(lambda=10)__pT1000_2000_unnorm.pdf.png") |
Edit | Attach | Topic revision: r2 - 2023-03-29 - DavideMelini
Home 
Plugins
Sandbox for tests
Support
Alice
Atlas
CMS
LHCb
Public Webs
Sandbox Web 
Create New Topic
Index
Search
Changes
Notifications
RSS Feed
Statistics
Account 
Copyright &© 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors. or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback