Deep-learned Top Tagging with a Lorentz Layer

Abstract

We introduce a new and highly efficient tagger for hadronically decaying top quarks, based on a deep neural network working with Lorentz vectors and the Minkowski metric. With its novel machine learning setup and architecture it allows us to identify boosted top quarks not only from calorimeter towers, but also including tracking information. We show how the performance of our tagger compares with QCD-inspired and image-recognition approaches and find that it significantly increases the performance for strongly boosted top quarks. Content

Deep-learning jets with uncertainties and more

“…3.2. Our toy BNN is trained to distinguish a public set of 600k top jet and QCD jet images each [10], which were generated with Pythia8 [27] for an LHC energy of 14 TeV and without pile-up or multiple interactions. As a simplified detector simulation we use Delphes [28] with the default ATLAS detector card.…”

“…Experimentally, it has the great advantage that we can train neural networks on mixed top pair events, where we first identify the leptonically decaying top quarks and then run the tagger on the hadronic recoil. Theoretically, the main features of top decays are safely perturbative [10], so unlike, for instance, in the case of quark-gluon separation [5] we do not expect detector effects and pile-up to have decisive impact on the tagging performance. On the other hand, detector effects and (related) systematic effects are going to be key factors in any application of machine learning techniques to subjet physics [14], especially if we eventually need to go beyond perfectly labelled actual jets to MC-enhanced training samples.…”

Mass agnostic jet taggers

“…The adversary is set to not be trainable, and the classifier weights are updated using the total loss of Eq. (8). However, only a small number of updates to the weights of the classifier are allowed.…”

FPGA-Accelerated Machine Learning Inference as a Service for Particle Physics Computing

“…In this case study, we consider the task of classifying collimated decays of top quarks in a jet from more common jets originating from lighter quarks or gluons. There are many ML approaches to this challenge in the literature [24] and a public dataset, developed from one of these studies, has been created for comparison [25,15]. The Pythia8 [26,27] generator is used to produce fully hadronic tt events for signal (known as"top quark jets") and QCD dijet events for background (known as "QCD jets") produced in 14 TeV proton-proton collisions.…”