Machine Learning in High Energy Physics Community White Paper May17,2019 Abstract:Machine learning has been appied problems in particle physics with nd training the article nh comm The mai objective of the doc ument is to connect and mot ivate th hese areas of research and development with the physics drivers of the High- will be of great benefit. Editors:Sergei Gleyzer,Paul Seyfert,Steven Schramm Contributors:Kim Albertsson,Piero Altoe2,Dustin Ande on,John Anderson,Michael Andrews,Juan Cala Laurent Wahid Bhim Bonac 1:9 Elias Coniav Kyle Cranm Claire Davids 13A1 A Giron 13.Pa Vava Gligon ar Tob Golling . on He 7, Gra Greenwood nas Hacker John Har vey Zahari Kassa Alexei Klim nio Limo Gilles Louppe17,Aa rita Mangu Pere Mato Helg Meinhard,Dario Menasce Lon ortg Manf -15 a 7,Ryan Ree ,Aurelius Rinkevicius Ed rdo amal Ror Aaron Sauers 13 Schwartzman Ho Mart 2 NVidia de mellon Unlversity LIP Lisb niversity of London ional Laboratory ,Bologna CERN University
Machine Learning in High Energy Physics Community White Paper May 17, 2019 Abstract: Machine learning has been applied to several problems in particle physics research, beginning with applications to high-level physics analysis in the 1990s and 2000s, followed by an explosion of applications in particle and event identification and reconstruction in the 2010s. In this document we discuss promising future research and development areas for machine learning in particle physics. We detail a roadmap for their implementation, software and hardware resource requirements, collaborative initiatives with the data science community, academia and industry, and training the particle physics community in data science. The main objective of the document is to connect and motivate these areas of research and development with the physics drivers of the High-Luminosity Large Hadron Collider and future neutrino experiments and identify the resource needs for their implementation. Additionally we identify areas where collaboration with external communities will be of great benefit. Editors: Sergei Gleyzer30, Paul Seyfert13, Steven Schramm32 Contributors: Kim Albertsson1 , Piero Altoe2 , Dustin Anderson3 , John Anderson4 , Michael Andrews5 , Juan Pedro Araque Espinosa6 , Adam Aurisano7 , Laurent Basara8 , Adrian Bevan9 , Wahid Bhimji10, Daniele Bonacorsi11 , Bjorn Burkle12, Paolo Calafiura10, Mario Campanelli9 , Louis Capps2 , Federico Carminati13, Stefano Carrazza13 , Yi-Fan Chen4 , Taylor Childers14, Yann Coadou15, Elias Coniavitis16, Kyle Cranmer17, Claire David18, Douglas Davis19, Andrea De Simone20, Javier Duarte21, Martin Erdmann22, Jonas Eschle23, Amir Farbin24, Matthew Feickert25, Nuno Filipe Castro6 , Conor Fitzpatrick26, Michele Floris13, Alessandra Forti27, Jordi Garra-Tico28 , Jochen Gemmler29, Maria Girone13, Paul Glaysher18, Sergei Gleyzer30, Vladimir Vava Gligorov31, Tobias Golling32, Jonas Graw2 , Lindsey Gray21, Dick Greenwood33, Thomas Hacker34, John Harvey13, Benedikt Hegner13, Lukas Heinrich17, Ulrich Heintz12, Ben Hooberman35, Johannes Junggeburth36, Michael Kagan37 , Meghan Kane38, Konstantin Kanishchev8 , Przemys law Karpi´nski13, Zahari Kassabov39, Gaurav Kaul40, Dorian Kcira3 , Thomas Keck29, Alexei Klimentov41, Jim Kowalkowski21, Luke Kreczko42, Alexander Kurepin43, Rob Kutschke21, Valentin Kuznetsov44, Nicolas K¨ohler36, Igor Lakomov13, Kevin Lannon45, Mario Lassnig13, Antonio Limosani46, Gilles Louppe17, Aashrita Mangu47, Pere Mato13, Helge Meinhard13, Dario Menasce48, Lorenzo Moneta13, Seth Moortgat49, Meenakshi Narain12, Mark Neubauer35, Harvey Newman3 , Sydney Otten50, Hans Pabst40, Michela Paganini51, Manfred Paulini5 , Gabriel Perdue21, Uzziel Perez52, Attilio Picazio53, Jim Pivarski54 , Harrison Prosper55, Fernanda Psihas56, Alexander Radovic57, Ryan Reece58, Aurelius Rinkevicius44, Eduardo Rodrigues7 , Jamal Rorie59, David Rousseau60, Aaron Sauers21, Steven Schramm32, Ariel Schwartzman37, Horst Severini61, Paul Seyfert13, Filip Siroky62, Konstantin Skazytkin43, Mike Sokoloff7 , Graeme Stewart63, Bob Stienen64, Ian Stockdale65, Giles Strong6 , Wei Sun4 , Savannah Thais51, Karen Tomko66, Eli Upfal12, Emanuele Usai12, Andrey Ustyuzhanin67, Martin Vala68, Sofia Vallecorsa69, Justin Vasel56, Mauro Verzetti70, Xavier Vilas´ıs-Cardona71, Jean-Roch Vlimant3 , Ilija Vukotic72, Sean-Jiun Wang30, Gordon Watts73, Michael Williams74 , Wenjing Wu75, Stefan Wunsch29, Kun Yang4 , Omar Zapata76 1 Lulea University of Technology 2 NVidia 3 California Institute of Technology 4 Google 5 Carnegie Mellon University 6 LIP Lisboa 7 University of Cincinnati 8 Universita e INFN, Padova 9 University of London 10 Lawrence Berkeley National Laboratory 11 Universita e INFN, Bologna 12 Brown University 13 CERN 1 arXiv:1807.02876v3 [physics.comp-ph] 16 May 2019
e National Laboratory 6 ronen-Synchrotror SISSA Trieste Italy Uni rsitat Z行rich Ecole Polytee dhalqueFedlo e de Lausann University of Illinois at Urbana-Champaign versity of Milan Brookhaven National Laboratory University of Bristo of Science Notre Dame University of Berkeley vert INFN,Mil c yofAmterdn Radboud Univerity Nmegc ton University 61n Rice Uni ty of Glasgo Ohio Sup ng-Wonju Nat 2
14 Argonne National Laboratory 15 CPPM Aix Marseille Univ CNRS/IN2P3 16 Universitaet Freiburg 17 New York University 18 Deutsches Elektronen-Synchrotron 19 Duke University 20 SISSA Trieste Italy 21 Fermi National Accelerator Laboratory 22 RWTH Aachen University 23 Universit¨at Z¨urich 24 University of Texas at Arlington 25 Southern Methodist University 26 Ecole Polytechnique Federale de Lausanne 27 University of Manchester 28 University of Cambridge 29 Karlsruher Institut f¨ur Technologie 30 University of Florida 31 LPNHE, Sorbonne Universit´e et Universit´e Paris Diderot, CNRS/IN2P3, Paris 32 Universit´e de Gen`eve 33 Louisiana Tech University 34 Purdue University 35 University of Illinois at Urbana-Champaign 36 Max Planck Institut f¨ur Physik 37 SLAC National Accelerator Laboratory 38 SoundCloud 39 University of Milan 40 Intel 41 Brookhaven National Laboratory 42 University of Bristol 43 Russian Academy of Sciences 44 Cornell University 45 University of Notre Dame 46 University of Melbourne 47 University of California Berkeley 48 Universita & INFN, Milano Bicocca 49 Vrije Universiteit Brussel 50 University of Amsterdam and Radboud University Nijmegen 51 Yale University 52 University of Alabama 53 University of Massachusetts 54 Princeton University 55 Florida State University 56 Indiana University 57 College of William and Mary 58 University of California, Santa Cruz 59 Rice University 60 Universite de Paris Sud 11 61 University of Oklahoma 62 Masaryk University 63 University of Glasgow 64 Radboud Universiteit Nijmegen 65 Altair Engineering 66 Ohio Supercomputer Center 67 Yandex School of Data Analysis 68 Technical University of Kosice 69 Gangneung-Wonju National University 70 University of Rochester 71 University of Barcelona 72 University of Chicago 73 University of Washington 74 Massachusetts Institute of Technology 2
3
75 Chinese Academy of Sciences 76 OProject and University of Antioquia 3
Contents 1 Preface 2 Introduction of Machine Lear ing Alg gorithms in HEp 6667 3 Machine Learning Applications and R&D Analysis and trigge 3.3Object Reconstruction.Identification.and Calibration 7788990 6 Matrix Element Machine Learning Method source Optimization and 4 Aca ademic Outreach and Engagement 4 13314455 5.2 I/O and Programming Languag traeteT ion Hardware 5.5.2D able HEP-NL a Formats Data Format Attributes 5.5.3 Interfaces and Middleware g 6 Computing and Hardware Resources 202 High Performance C omputing 66 Opportunistic resources 222 6.7 Data Storage and Availability 6 Machine Leamni 222222 7 Training the community 22 223 9 Conclusions 10 Acknowledgements 4
Contents 1 Preface 6 2 Introduction 6 2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2 Brief Overview of Machine Learning Algorithms in HEP . . . . . . . . . . . . . . . . . . . . . . . 6 2.3 Structure of the Document . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3 Machine Learning Applications and R&D 7 3.1 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3.2 Real Time Analysis and Triggering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 3.3 Object Reconstruction, Identification, and Calibration . . . . . . . . . . . . . . . . . . . . . . . . 8 3.4 End-To-End Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 3.5 Sustainable Matrix Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 3.6 Matrix Element Machine Learning Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 3.7 Learning the Standard Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 3.8 Theory Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.9 Uncertainty Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.10 Monitoring of Detectors, Hardware Anomalies and Preemptive Maintenance . . . . . . . . . . . . 13 3.11 Computing Resource Optimization and Control of Networks and Production Workflows . . . . . 13 4 Collaborating with other communities 13 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4.2 Academic Outreach and Engagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 4.3 Machine Learning Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 4.4 Collaborative Benchmark Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 4.5 Industry Engagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 4.6 Machine Learning Community-at-large Outreach . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 5 Machine Learning Software and Tools 16 5.1 Software Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 5.2 I/O and Programming Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 5.3 Software Interfaces to Acceleration Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 5.4 Parallelization and Interactivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 5.5 Internal and External ML tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 5.5.1 Machine Learning Data Formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 5.5.2 Desirable HEP-ML Software and Data Format Attributes . . . . . . . . . . . . . . . . . . 19 5.5.3 Interfaces and Middleware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 6 Computing and Hardware Resources 19 6.1 Resource Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 6.2 Graphical Processing Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 6.3 Cloud TPUs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 6.4 High Performance Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 6.5 Field Programmable Gate Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 6.6 Opportunistic Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 6.7 Data Storage and Availability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 6.8 Software Distribution and Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 6.9 Machine Learning As a Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 7 Training the community 22 8 Roadmap 22 8.1 Timeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 8.2 Steps to Deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 9 Conclusions 23 10 Acknowledgements 23 4
Element Mfetbods
A Appendix 24 A.1 Matrix Element Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 5