Abstract
We apply computer vision with deep learning — in the form of a convolutional neural network (CNN) — to build a highly effective boosted top tagger. Previous work (the “DeepTop” tagger of Kasieczka et al) has shown that a CNN-based top tagger can achieve comparable performance to state-of-the-art conventional top taggers based on high-level inputs. Here, we introduce a number of improvements to the DeepTop tagger, including architecture, training, image preprocessing, sample size and color pixels. Our final CNN top tagger outperforms BDTs based on high-level inputs by a factor of ∼ 2-3 or more in background rejection, over a wide range of tagging efficiencies and fiducial jet selections. As reference points, we achieve a QCD background rejection factor of 500 (60) at 50% top tagging efficiency for fully-merged (non-merged) top jets with pT in the 800-900 GeV (350-450 GeV) range. Our CNN can also be straightforwardly extended to the classification of other types of jets, and the lessons learned here may be useful to others designing their own deep NNs for LHC applications.
Highlights
To play a key role in solutions to the hierarchy problem, and they can naturally produce boosted top quarks in their decays
Our final convolutional neural network (CNN) top tagger outperforms boosted decision trees (BDTs) based on high-level inputs by a factor of ∼ 2–3 or more in background rejection, over a wide range of tagging efficiencies and fiducial jet selections
Our CNN can be straightforwardly extended to the classification of other types of jets, and the lessons learned here may be useful to others designing their own deep neural networks (NNs) for LHC applications
Summary
The fat jets used in this paper are taken from all-hadronic ttand QCD dijet events generated in proton-proton collisions using Pythia 8.219 [49], where multiparton interactions and pileup are turned off for simplicity. As discussed in the Introduction, we will study improvements to the DeepTop tagger using two very different samples of jet images. For the CMS sample, we will consider both a cut-based tagger that combines the HTTV2 variables with the Nsubjettiness variable τ3/τ2 (motivated by the recent CMS note on top tagging [42]), as well as a BDT trained on these variables. For the former, we varied simple window cuts on each of the variables, as in [42]. We will consider a number of improvements to the DeepTop tagger that, taken together, demonstrate for the first time that CNNs can significantly outperform conventional taggers
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
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 call
