QCD-aware recursive neural networks for jet physics

Gilles Louppe, Kyunghyun Cho, Cyril Becot, Kyle Cranmer

Research output: Contribution to journalArticle

Abstract

Recent progress in applying machine learning for jet physics has been built upon an analogy between calorimeters and images. In this work, we present a novel class of recursive neural networks built instead upon an analogy between QCD and natural languages. In the analogy, four-momenta are like words and the clustering history of sequential recombination jet algorithms is like the parsing of a sentence. Our approach works directly with the four-momenta of a variable-length set of particles, and the jet-based tree structure varies on an event-by-event basis. Our experiments highlight the flexibility of our method for building task-specific jet embeddings and show that recursive architectures are significantly more accurate and data efficient than previous image-based networks. We extend the analogy from individual jets (sentences) to full events (paragraphs), and show for the first time an event-level classifier operating on all the stable particles produced in an LHC event.

Original languageEnglish (US)
Article number57
JournalJournal of High Energy Physics
Volume2019
Issue number1
DOIs
StatePublished - Jan 1 2019

Fingerprint

quantum chromodynamics
physics
sentences
momentum
machine learning
classifiers
embedding
calorimeters
flexibility
histories

Keywords

  • Jets
  • QCD Phenomenology

ASJC Scopus subject areas

  • Nuclear and High Energy Physics

Cite this

QCD-aware recursive neural networks for jet physics. / Louppe, Gilles; Cho, Kyunghyun; Becot, Cyril; Cranmer, Kyle.

In: Journal of High Energy Physics, Vol. 2019, No. 1, 57, 01.01.2019.

Research output: Contribution to journalArticle

Louppe, Gilles ; Cho, Kyunghyun ; Becot, Cyril ; Cranmer, Kyle. / QCD-aware recursive neural networks for jet physics. In: Journal of High Energy Physics. 2019 ; Vol. 2019, No. 1.
@article{21c067905dd24433b29fe26d35928f0e,
title = "QCD-aware recursive neural networks for jet physics",
abstract = "Recent progress in applying machine learning for jet physics has been built upon an analogy between calorimeters and images. In this work, we present a novel class of recursive neural networks built instead upon an analogy between QCD and natural languages. In the analogy, four-momenta are like words and the clustering history of sequential recombination jet algorithms is like the parsing of a sentence. Our approach works directly with the four-momenta of a variable-length set of particles, and the jet-based tree structure varies on an event-by-event basis. Our experiments highlight the flexibility of our method for building task-specific jet embeddings and show that recursive architectures are significantly more accurate and data efficient than previous image-based networks. We extend the analogy from individual jets (sentences) to full events (paragraphs), and show for the first time an event-level classifier operating on all the stable particles produced in an LHC event.",
keywords = "Jets, QCD Phenomenology",
author = "Gilles Louppe and Kyunghyun Cho and Cyril Becot and Kyle Cranmer",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/JHEP01(2019)057",
language = "English (US)",
volume = "2019",
journal = "Journal of High Energy Physics",
issn = "1126-6708",
publisher = "Springer Verlag",
number = "1",

}

TY - JOUR

T1 - QCD-aware recursive neural networks for jet physics

AU - Louppe, Gilles

AU - Cho, Kyunghyun

AU - Becot, Cyril

AU - Cranmer, Kyle

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Recent progress in applying machine learning for jet physics has been built upon an analogy between calorimeters and images. In this work, we present a novel class of recursive neural networks built instead upon an analogy between QCD and natural languages. In the analogy, four-momenta are like words and the clustering history of sequential recombination jet algorithms is like the parsing of a sentence. Our approach works directly with the four-momenta of a variable-length set of particles, and the jet-based tree structure varies on an event-by-event basis. Our experiments highlight the flexibility of our method for building task-specific jet embeddings and show that recursive architectures are significantly more accurate and data efficient than previous image-based networks. We extend the analogy from individual jets (sentences) to full events (paragraphs), and show for the first time an event-level classifier operating on all the stable particles produced in an LHC event.

AB - Recent progress in applying machine learning for jet physics has been built upon an analogy between calorimeters and images. In this work, we present a novel class of recursive neural networks built instead upon an analogy between QCD and natural languages. In the analogy, four-momenta are like words and the clustering history of sequential recombination jet algorithms is like the parsing of a sentence. Our approach works directly with the four-momenta of a variable-length set of particles, and the jet-based tree structure varies on an event-by-event basis. Our experiments highlight the flexibility of our method for building task-specific jet embeddings and show that recursive architectures are significantly more accurate and data efficient than previous image-based networks. We extend the analogy from individual jets (sentences) to full events (paragraphs), and show for the first time an event-level classifier operating on all the stable particles produced in an LHC event.

KW - Jets

KW - QCD Phenomenology

UR - http://www.scopus.com/inward/record.url?scp=85059835473&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85059835473&partnerID=8YFLogxK

U2 - 10.1007/JHEP01(2019)057

DO - 10.1007/JHEP01(2019)057

M3 - Article

AN - SCOPUS:85059835473

VL - 2019

JO - Journal of High Energy Physics

JF - Journal of High Energy Physics

SN - 1126-6708

IS - 1

M1 - 57

ER -