ONLINE: Workshop on Perspectives and Applications of Deep Learning for Accelerated Scientific Discovery in Physics


Break-throughs in deep learning (DL) technology are revolutionising the amount, rate and nature of information that can be extracted from data, creating new opportunities to accelerate scientific discovery in ways that have never been imagined before.

Billedresultat for machine learning workshop

This workshop is a follow up of the previous one organised on September 2019 at the NBI. During these two days on May 2020, we shall  again focus on two topical themes on DL Accelerated Discovery.

• The use of DL architectures to generate unbiased and objective models for analysis of experimental data.

• The use of DL methods for optimising data collection protocols in real time and for discovering anomalies in experimental data and enhancing the efficiency of the scientific workflow.

These intersections of scientific discovery and DL science are unique and the workshop will now explore potential research avenues to understand how to best develop these technologies to accelerate discovery in Physics.

The program for the workshop will consist of plenary sessions followed by panel discussions (two panels each day), and final close-out.

On Thursday 14th morning, plenary seminars will review the state of the art in DL and how it is currently used in selected areas of Physics, as well as on the challenges that these experimental techniques face that can be addressed by DL.

On Thursday 14th afternoon, panel groups, led by a convener, will explore and report in how DL methods can become more generically applicable to a wide range of  experiments, in the areas of Experiment design, Data collection and instrumentation control, Data reduction Analysis, and extraction of science.            

On Friday 15th morning panel groups, led by a convener,  will explore techniques, one panel focusing on (un)supervised and reinforcement learning, and the other on Deep Artificial Neural Networks.

On Friday 15th afternoon, panel conveners will present the outcomes of the panel discussions and the meeting will close with a discussion and presentation of an overall summary. A tentative roadmap on how to make progress in the application of DL for Accelerated Scientific Discovery in Physics will also be proposed.

Looking very much forward to seeing you in  May!



  • Adriano Agnello
  • Alberto Nannarelli
  • Alejandro Vigna Gómez
  • Alessandra Camplani
  • Amira Moussa
  • Anasua Chatterjee
  • Anders Dahl
  • Anders Kringhøj
  • Andrea Kirsch
  • Andreas Salzburger
  • Andy Sode Anker
  • Anne Klitsch
  • Birgitte Nilsson
  • Bo Milvang-Jensen
  • Carlo Cannarozzo
  • Celine Durniak
  • Christa Gall
  • Christian Jespersen
  • Christine Rasmussen
  • Deividas Sabonis
  • Dina Rapp
  • Emil Thyge Skaaning Kjær
  • Evgenii Velichko
  • Ferdinand Kuemmeth
  • Gagik Vardanyan
  • Gerald Kneller
  • Giacomo Girelli
  • Haixing Fang
  • Heloisa Bordallo
  • Heorhii Bohuslavskyi
  • Hugo Pfister
  • I-Ju Chen
  • Iary Davidzon
  • John Weaver
  • Jos Cooper
  • Juan Carlos Estrada Saldaña
  • Juan Manuel Carmona Loaiza
  • Kenneth Skovhede
  • Lamar Moore
  • Luca Izzo
  • Maher Sahyoun
  • Mark Hagen
  • Markus Ahlers
  • Martin Hoffmann Petersen
  • Martin Nors Pedersen
  • Michaela Hirschmann
  • Milena Bajic
  • Monica-Elisabeta Lacatusu
  • Nicolas Magnard
  • Nikki Arendse
  • Nina Bonaventura
  • Oleg Ruchayskiy
  • Oswin Krause
  • Petr Sittner
  • Rasmus Schlosser
  • Shashank Shalgar
  • Sofie Bruun
  • Steen Hansen
  • stefania xella
  • Toby Perring
  • Torbjørn Rasmussen
  • Troels Petersen
  • Zoe Ansari
    • 1
      Speaker: Heloisa Bordallo (NBI)
    • 2
      Machine learning and particle physics
      Speaker: Andreas Salzburger (CERN)
    • 3
      Data Challenges for accelerating scientific discovery at Neutron facilities
      Speaker: Jonathan Taylor (ess)
    • 10:45 AM
      coffe break
    • 4
      Collaboration on analysis of 3D image data the QIM Center
      Speaker: Anders Dahl (Technical University of Denmark)
    • 5
      A data-driven approach in Astrophysics
      Speaker: Iary Davidzon (NBI/Cosmic Dawn Center)
    • 12:45 PM
      lunch break
    • PS1: Themes: ML in hardware, and in Experiment control and design. Example of question to be debated: Making decisions in the flight in how to change instrument configuration

      Making decisions in flight in how to change instrument configuration

    • PS2: Themes: data collection, reduction and analysis. Example of topics of discussion : 1) Exploring noisy data and analyzing large amount of data [ Minimizing the reliance on good training data and biases from training on simulated data] 2) Changes in data taking/content/quality [dealing with experimental uncertainties]
    • PS3: Techniques: Unsupervised/reinforced learning. Example of topics to discuss: Structure in running conditions for good/bad data, Outlier detection, Checking if background has several components. Unsupervised learning : remove vs add bias, in organizing information.
    • PS4: Techniques: Deep NN. Examples of topics to discuss: Finding optimum in very high dimensional spaces, Finding simpler methods that will work with fewer parameters under many conditions
    • 11:30 AM
      lunch break
    • 6
      Report from ML in hardware, and in Experiment control and design
      Speaker: Alberto Nannarelli (DTU)
    • 7
      Report from data collection, reduction and analysisRe
      Speaker: Lamar Moore (STFC)
    • 8
      Report from: Techniques: Unsupervised/reinforced learning
      Speaker: Iary Davidzon (NBI/Cosmic Dawn Center)
    • 9
      Report from Techniques: Deep NN
      Speaker: Troels Petersen (Niels Bohr Institute)
    • 10
      Wrap up and ideas for a roadmap
      Speakers: Heloisa Bordallo (NBI), Troels Petersen (Niels Bohr Institute)