Learning to Discover : Advanced Pattern Recognition

Europe/Zurich
Institut Pascal, Orsay. Paris

Institut Pascal, Orsay. Paris

Andreas Salzburger (CERN), David Rousseau (LAL-Orsay, FR), Sabrina Soccard (Institut Pascal)
Description

Pattern recognition is the fundament of the event reconstruction in high energy physics, but has also become - alongside with the detector simulation - one of the main computing intensive tasks.

For the Large Hadron Collider (LHC), the  luminosity is expected to increase by a factor five by 2025, leading to proton collision with very large complexity. The current pattern recognition algorithms, which role is to reconstruct the individual particles from the energy deposition on the detectors, are stretching to their limits. A general effort is needed by putting together experts from the different experiments, as well as Data Science experts in novel optimisation algorithms, to try out completely different approaches and foster from recent advances in machine learning: Graph Networks, Recurrent Neural Networks, Convolutional Neural Networks, Monte Carlo Tree Search, Probabilistic hashing are methods which have been proposed to solve this problem, on the one hand, and further optimisation of current pattern recognition algorithms, such as concurrent algorithmic execution or optimisation for accelerated hardware are pursued on the other hand. Several programs exploiting machine learning techniques have been based lately within the context of the tracking machine learning challenge, in the field of quantum computing and carried out by HEP experiments in order to meet the requirements for future data taking campaigns at the LHC and other HEP experiments. 

Link to Institut Pascal web page.

 

Participants
  • Aishik Ghosh
  • Amélie Chatelain
  • Anais Möller
  • Andreas Salzburger
  • Biswas Biswajit
  • Corentin Allaire
  • David Rousseau
  • Edward Moyse
  • Emilio Cortina Labra
  • Eric Chabert
  • Fabian Klimpel
  • Felice Pantaleo
  • Habrard Amaury
  • Iacopo Poli
  • Jean-Roch Vlimant
  • Jessica Leveque
  • Jonathan Shlomi
  • Kazuya Koyama
  • Laurent Basara
  • Laurent Daudet
  • Lucy Linder
  • Marcel Kunze
  • Marco Rovere
  • Markus Elsing
  • Moritz Kiehn
  • Sabrina Amrouche
  • Sazonov Vasily
  • Sergey Gorbunov
  • Slava Voloshynovskiy
  • Talbot Hugues
  • Tobias Isenberg
  • Vladimir Gligorov
  • Xiayo Wang
    • 09:30 10:30
      Introduction, Summary: Introduction

      Initial registration
      Introduction

    • 10:30 11:00
      Coffee break: Coffee break 1
    • 11:00 12:30
      Scientific Session: Pattern recognition in HEP

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Markus Elsing (CERN)
      • 11:00
        Pattern Recognition in HEP 1h

        Review and outlook on track reconstruction & pattern recognition in high energy physics

        Speaker: Markus Elsing (CERN)
    • 12:30 14:00
      Lunch: Monday Lunch
    • 14:00 15:30
      Scientific Session: Pattern recognition

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Prof. Talbot Hugues (CentraleSupélec, Université Paris-Saclay)
      • 14:00
        Path operators for thin objects restoration 1h

        In this talk we will present path operators, which are efficient recursive mathematical morphology connected operators that use paths as structuring elements. These operators are designed to preserve thin objects in images, such as hair, cilia, vessels, oriented textures, etc, which are traditionally very difficult to filter using classical operators in many settings. By combining these filters, we show how we can propose a vesselness operator with significant better performance than the traditional linear operators based on the Hessian (Frangi, Sato, etc) or the structure tensor. We also show recent work on how to use these operators as regularizers in variational frameworks for image restoration, in the context of discrete calculus.

        Speaker: Prof. Talbot Hugues (CentraleSupélec, Université Paris-Saclay)
    • 15:30 16:00
      Coffee break: Coffee break 2
    • 16:00 17:30
      Discussion, Collaboration: Discussions
    • 09:20 09:30
      Announcement 10m
      Speaker: Andreas Salzburger (CERN)
    • 09:30 11:00
      Scientific Session: Summary of the Tracking ML Challenge

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Moritz Kiehn (Universite de Geneve (CH))
      • 09:30
        Summary of the TrackML Challenge 1h
        Speaker: Moritz Kiehn (Universite de Geneve (CH))
      • 10:30
        Q&A, Discussion 30m
    • 11:00 11:30
      Coffee break: Coffee break 3
    • 11:30 13:00
      Scientific Session: Track reconstruction with similarity hashing

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Sabrina Amrouche (Université de Geneve (CH))
      • 11:30
        Track reconstruction with similarity hashing 1h
        Speaker: Sabrina Amrouche (Université de Geneve (CH))
      • 12:30
        Q&A, Discussion 30m
    • 13:00 14:00
      Lunch: Tuesday lunch
    • 14:00 15:30
      Scientific Session: Graph networks for track reconstruction

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Marcel Kunze (Heidelberg University)
      • 14:00
        Graph networks for track reconstruction 1h
        Speaker: Marcel Kunze (Heidelberg University)
      • 15:00
        Q&A, Discussion 30m
    • 15:30 16:00
      Coffee break: Coffee break 4
    • 16:00 17:30
      Discussion, Collaboration: Discussions
    • 17:30 18:30
      Social: Science Hour
    • 09:30 11:00
      Scientific Session: GPU usage in HEP

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Conveners: Felice Pantaleo (CERN), Marco Rovere (CERN)
    • 11:00 11:30
      Coffee break: Coffee break 5
    • 11:30 13:00
      Scientific Session: Variable latency tracking at 30 MHz

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Vladimir Gligorov (Centre National de la Recherche Scientifique (FR))
      • 11:30
        Variable latency tracking at 30 MHz 1h
        Speaker: Vladimir Gligorov (Centre National de la Recherche Scientifique (FR))
      • 12:30
        Q&A, Discussion 30m
    • 13:00 14:00
      Lunch: Wednesday lunch
    • 14:00 15:30
      Scientific Session: Performance vs Portability vs Productivity vs Precision : a trail in the hardware and software jungle.

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: David Chamont (LAL)
      • 14:00
        Performance vs Portability vs Productivity vs Precision : a trail in the hardware and software jungle. 1h
        Speaker: David Chamont (LAL)
      • 15:00
        Q&A, Discussion 30m
    • 15:30 16:00
      Coffee break: Coffee break 6
    • 16:00 17:30
      Discussion, Collaboration: Discussions
    • 09:30 11:00
      Scientific Session: Quantum Computing meets HEP

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Lucy Linder (Lawrence Berkeley National Laboratory)
      • 09:30
        Quantum Computing meets HEP 1h
        Speaker: Lucy Linder
      • 10:30
        Q&A, Discussion 30m
    • 11:00 11:30
      Coffee break: Coffee break 7
    • 11:30 13:00
      Scientific Session: Introduction to random matrices and tensors, and application to Principal Component Analysis

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Vasily Sazonov (U)
    • 13:00 14:00
      Lunch: Thursday lunch
    • 14:00 15:30
      Scientific Session: Optical Processing Unit

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Laurent Daudet
      • 14:00
        Optical random features for large-scale machine learning 45m
        Speaker: Laurent Daudet
      • 14:45
        Demo of OPU on ML applications 45m
        Speaker: Iacopo Poli
    • 15:30 16:00
      Coffee break: Coffee break 8
    • 16:00 17:30
      Hands-on session: First attempts of using OPU on HEP problems
      Conveners: Aishik Ghosh (Centre National de la Recherche Scientifique (FR)), Biswajit Biswas (LAL-Orsay), David Rousseau (LAL-Orsay, FR), Dr Laurent Basara (LAL/LRI, Université Paris Saclay)
    • 19:00 21:00
      Social: 1st week social dinner at the Gramophone
    • 09:30 11:00
      Scientific Session: Event display in HEP

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Edward Moyse (University of Massachusetts (US))
    • 11:00 11:30
      Coffee break: Coffee break 9
    • 11:30 13:00
      Scientific Session: Virtual Reality and Augmented Reality in HEP

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Isenberg Tobias
      • 11:30
        VR and AR in HEP 1h
        Speaker: Tobias Isenberg (Inria)
      • 12:30
        Q&A, Discussion 30m
    • 13:00 14:00
      Lunch: Friday lunch
    • 14:00 15:30
      Hands-on session: Hands-on: Phoenix, run and develop
      Convener: Emilio Cortina
      • 14:00
        Phoenix Hands-on 1h
        Speaker: Emilio Cortina
    • 09:30 09:45
      Introduction, Summary: Introduction - week 2

      Initial registration
      Introduction

      • 09:30
        Institut Pascal - an introduction 15m
        Speaker: Jacques Bittoun
    • 09:45 11:15
      Scientific Session: HEP.TrkX, EXA.TrkX and beyond

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Dr Jean-Roch Vlimant (California Institute of Technology (US))
      • 09:45
        HEP.TrkX, Exa.TrkX & beyond 1h
        Speaker: Jean-Roch Vlimant (California Institute of Technology (US))
      • 10:45
        Q&A, Discussion 30m
    • 11:15 11:30
      Coffee break: Coffee break 10
    • 11:30 13:00
      Scientific Session: Scattering transform for pattern recognition

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Edouard Oyallon
      • 11:30
        Scattering transform for pattern recognition 1h
        Speaker: Edouard Oyallon (Centre de Vision Numérique)
      • 12:30
        Q&A, Discussion 30m
    • 13:00 14:00
      Lunch: Monday lunch
    • 14:00 15:30
      Scientific Session: Detector and Algorithm development with ACTS

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Andreas Salzburger (CERN)
    • 15:30 16:00
      Coffee break: Coffee break 11
    • 16:00 17:00
      Hands-on session: Detector and Algorithm development with ACTS
      Convener: Andreas Salzburger (CERN)
      • 16:00
        ACTS hands-on 1h
        Speaker: Andreas Salzburger (CERN)
    • 09:30 11:00
      Scientific Session: A classical fast tracking algorithm: Mikado

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Sergey Gorbunov (Johann-Wolfgang-Goethe Univ. (DE))
      • 09:30
        The Mikado algorithm & beyond 1h
        Speaker: Sergey Gorbunov (Johann-Wolfgang-Goethe Univ. (DE))
      • 10:30
        Q&A, Discussion 30m
    • 11:00 11:15
      Coffee break: Coffee break 12
    • 11:15 12:45
      Scientific Session: Timing in pattern and event reconstruction

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Corentin Allaire (Centre National de la Recherche Scientifique (FR))
      • 11:15
        Timing detectors in Phase-2 30m
        Speaker: Corentin Allaire (Centre National de la Recherche Scientifique (FR))
      • 11:45
        Timing in propagation and pattern recognition 30m
        Speaker: Fabian Klimpel (Max-Planck-Institut fur Physik (DE))
      • 12:15
        Q&A, Discussion 30m
    • 12:45 13:45
      Lunch: Tuesday lunch
    • 17:00 18:00
      Scientific hour == drinks ! 1h

      In the "cathedral" on first floor

    • 09:30 10:45
      HEP meets ASTRO: Running Data Challenges
      Convener: David Rousseau (LAL-Orsay, FR)
    • 10:45 11:45
      HEP meets ASTRO: N-Body simulations and beyond
      Convener: Kazuya Koyama (University of Portsmouth)
      • 10:45
        N-Body simulations and beyond 45m
        Speaker: Kazuya Koyama (University of Portsmouth)
      • 11:30
        Q&A, Discussion 15m
    • 11:45 12:00
      Coffee break: HEP meets ASTRO coffee 1
    • 12:00 13:15
      HEP meets ASTRO: Metric Learning
      Convener: Amaury Habrard
      • 12:00
        Metric learning 1h
        Speaker: Prof. Amaury Habrard (Université Saint-Étienne)
      • 13:00
        Q&A, Discussion 15m
    • 13:15 14:15
      Lunch: HEP meets ASTRO buffet
    • 14:15 15:30
      HEP meets ASTRO: Deep learning for Cosmology
      Convener: Anais Moller
    • 15:30 16:00
      Coffee break: HEP meets ASTRO coffee 2
    • 16:00 17:30
      Discussion, Collaboration: HEP meets ASTRO discussion
    • 19:30 21:30
      Social: Conference dinner at "La part des anges"
    • 10:30 12:00
      Scientific Session: Fast indexing based on the latent space representation

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Slava Voloshynovskiy
      • 10:30
        Fast indexing based on the latent space representation 1h
        Speaker: Slava Voloshynovskiy (Université de Genève)
      • 11:30
        Q&A, Discussion 30m
    • 12:00 13:30
      Lunch: Thursday lunch
    • 13:30 15:00
      Scientific Session: Graph networks for vertexing

      Standard scientific session.
      Format: 45 minutes contribution + 45 minutes discussion

      Convener: Jonathan Shlomi (Weizmann Institute of Science (IL))
      • 13:30
        Graph networks for vertex reconstruction 1h
        Speaker: Jonathan Shlomi (Weizmann Institute of Science (IL))
      • 14:30
        Q&A, Discussion 30m
    • 15:00 15:30
      Coffee break: Coffee break 13
    • 09:30 10:30
      Introduction, Summary: An attempted summary

      Initial registration
      Introduction

      Convener: Andreas Salzburger (CERN)
    • 10:30 11:00
      Coffee break: Coffee break 14
    • 11:00 12:00
      Discussion, Collaboration: Springer editor team discussion (t.b.c.)