QCD-Aware Recursive Neural Networks for Jet Physics

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.

Note:

16 pages, 5 figures, 3 appendices, corresponding code at https://github.com/glouppe/recnn

Jets
QCD Phenomenology
quantum chromodynamics
neural network
numerical methods
numerical calculations
CERN LHC Coll
track data analysis: jet
shape analysis: jet
jet: transverse momentum

References(64)

Figures(3)

[1]

Jet-Images: Computer Vision Inspired Techniques for Jet Tagging

- JHEP 02 (2015) 118
•
e-Print:
- 1407.5675
•
DOI:
- 10.1007/JHEP02(2015)118

[2]

Jet-images — deep learning edition

Luke de Oliveira
(
- Stanford U., Math. Dept.
)
,
Michael Kagan
(
- SLAC
)
,
Lester Mackey
(
- Stanford U., Statistics Dept.
)
,
Benjamin Nachman
(
- SLAC
)
,
Ariel Schwartzman
(
- SLAC
)

- JHEP 07 (2016) 069
•
e-Print:
- 1511.05190
•
DOI:
- 10.1007/JHEP07(2016)069

[3]

Playing Tag with ANN: Boosted Top Identification with Pattern Recognition

Leandro G. Almeida
(
- Ecole Normale Superieure, CNRS
)
,
Mihailo Backović
(
- Louvain U., CP3
)
,
Mathieu Cliche
(
- Cornell U., LEPP
)
,
Seung J. Lee
(
- KAIST, Taejon and
- Korea Inst. Advanced Study, Seoul
)
,
Maxim Perelstein
(
- Cornell U., LEPP
)

- JHEP 07 (2015) 086
•
e-Print:
- 1501.05968
•
DOI:
- 10.1007/JHEP07(2015)086

[4]

Jet Substructure Classification in High-Energy Physics with Deep Neural Networks

Pierre Baldi
(
- UC, Irvine
)
,
Kevin Bauer
(
- UC, Irvine
)
,
Clara Eng
(
- UC, Berkeley, Chem. Eng. Dept.
)
,
Peter Sadowski
(
- UC, Irvine
)
,
Daniel Whiteson
(
- UC, Irvine
)

- Phys.Rev.D 93 (2016) 9, 094034
•
e-Print:
- 1603.09349
•
DOI:
- 10.1103/PhysRevD.93.094034

[5]

Parton Shower Uncertainties in Jet Substructure Analyses with Deep Neural Networks

- Phys.Rev.D 95 (2017) 1, 014018
•
e-Print:
- 1609.00607
•
DOI:
- 10.1103/PhysRevD.95.014018

[6]

Deep learning in color: towards automated quark/gluon jet discrimination

Patrick T. Komiske
(
- MIT, Cambridge, CTP
)
,
Eric M. Metodiev
(
- MIT, Cambridge, CTP
)
,
Matthew D. Schwartz
(
- Harvard U., Phys. Dept. and
- Harvard U.
)

- JHEP 01 (2017) 110
•
e-Print:
- 1612.01551
•
DOI:
- 10.1007/JHEP01(2017)110

[7]

Deep-learning Top Taggers or The End of QCD?

- JHEP 05 (2017) 006
•
e-Print:
- 1701.08784
•
DOI:
- 10.1007/JHEP05(2017)006

[8]

Jet Flavor Classification in High-Energy Physics with Deep Neural Networks

Daniel Guest
(
- UC, Irvine
)
,
Julian Collado
(
- UC, Irvine
)
,
Pierre Baldi
(
- UC, Irvine
)
,
Shih-Chieh Hsu
(
- Washington U., Seattle
)
,
Gregor Urban
(
- UC, Irvine
)

et al.

- Phys.Rev.D 94 (2016) 11, 112002
•
e-Print:
- 1607.08633
•
DOI:
- 10.1103/PhysRevD.94.112002

[9]

Learning task-dependent distributed representations by backpropagation through structure / Int. Conf

C. Goller
,
A. Kuchler

- Neural Networks 1 (1996) 347

[10]

Parsing natural scenes and natural language with recursive neural networks, in Proceedings of the 28th international conference on machine learning (ICML-11), Bellevue U.S.A / pg. 129

R. Socher
,
C.C. Lin
,
C. Manning
,
A.Y. Ng

[11]

Semi-supervised recursive autoencoders for predicting sentiment distributions, in Proceedings of the Conference on Empirical Methods in Natural Language Processing, Edinburgh U.K / pg. 151

R. Socher
,
J. Pennington
,
E.H. Huang
,
A.Y. Ng
,
C.D. Manning

[12]

On the Properties of Neural Machine Translation: Encoder-Decoder Approaches

e-Print:
- 1409.1259

[13]

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

et al.

e-Print:
- 1406.1078

[14]

Sentence modeling with gated recursive neural network, in Proceedings of theConference on Empirical Methods in Natural Language Processing, Lisbon Portugal , pg. 793

X. Chen
,
X. Qiu
,
C. Zhu
,
S. Wu
,
X. Huang

[15]

Deep Learning, MIT Press, Cambridge U.S.A

I. Goodfellow
,
Y. Bengio
,
A. Courville

[16]

Longitudinally invariant $K_t$ clustering algorithms for hadron hadron collisions

S. Catani
(
- CERN
)
,
Yuri L. Dokshitzer
(
- St. Petersburg, INP and
- Lund U.
)
,
M.H. Seymour
(
- Lund U.
)
,
B.R. Webber
(
- CERN
)

- Nucl.Phys.B 406 (1993) 187-224
•
DOI:
- 10.1016/0550-3213(93)90166-M

[17]

Better jet clustering algorithms

- JHEP 08 (1997) 001
•
e-Print:
- hep-ph/9707323
•
DOI:
- 10.1088/1126-6708/1997/08/001

[18]

The anti- $k_t$ jet clustering algorithm

- JHEP 04 (2008) 063
•
e-Print:
- 0802.1189
•
DOI:
- 10.1088/1126-6708/2008/04/063

[19]

Towards Jetography

Gavin P. Salam
(
- Paris, LPTHE
)

- Eur.Phys.J.C 67 (2010) 637-686
•
e-Print:
- 0906.1833
•
DOI:
- 10.1140/epjc/s10052-010-1314-6

[20]

Rectified linear units improve restricted boltzmann machines, in Proceedings of the 27th international conference on machine learning (ICML-10), Haifa Israel , pg. 807