3rd Workshop on Graph Neural Networks for Neutrino Telescope Event Reconstruction (GraphNeT III)

Europe/Copenhagen
Strandhotellet, Sandvig

Strandhotellet, Sandvig

Strandhotellet Strandpromenaden 7, Sandvig Bornholm, Denmark
Andreas Søgaard, Philipp Eller (TUM), Rasmus Ørsøe (NBI), Troels Petersen (Niels Bohr Institute)
Description

What is it about?

This is the third workshop dedicated to GraphNeT – Graph neural networks for neutrino telescope event reconstruction. The goal is to bring together researchers working at the intersection of neutrino telescope experiments and machine learning to meet likeminded researchers, discuss the latest progress, and develop new solutions to physics challenges using graph neural networks (GNNs) — putting graphs to work.  

The workshop, and particularly the two half-day hackathons, are focussed on putting the GraphNeT framework into use for physics — in IceCube and other experiments. In this way, we hope to foster collaboration on common tools in order to advance physics research faster than individual experiments can on their own. 

Who is it for?

The workshop is aimed at active practitioners who are eager to learn about GNNs, to showcase their existing physics work leveraging GNNs, and to participate in active, hands-on problem solving using GNNs. Participants who may not have substantial experience with GNNs, or machine learning (ML) in general, but who are interested in leveraging these techniques in their physics analysis are encouraged to join. As are ML experts, keen on exploring the newest technical developments within the field and help ensure their impact on neutrino physics. 

While the workshop's presentations can feature most work at the intersection of ML and neutrino telescopes, hackathon participants ought to have an interest in trying out the GraphNeT code — either because they’re open to using it as their main framework, or because they want to see how other people are working with GNNs.  We will have some experiment-independent MC data, so people should be able to show up with nothing more than their laptop and still be able to do some actual work/experimentation.

Where is it held?

The workshop will be held on scenic Bornholm, a Danish island in the Baltic Sea. Specifically, the workshop activities will take place at Strandhotellet, a hotel and conference venue at the island's northernmost tip. 

There may be some limitations to the workshop's capacity, in which case preference will be given to attendees who are actively working at the intersection of machine learning and neutrino telescope physics.

Funding

The organisers would like to acknowledge the generous support for this workshop, provided by the Danish Data Science Academy.

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 890778.

Participants
28
    • 08:30 09:00
      Arrival and coffee 30m
    • 09:00 10:30
      Presentations: Tuesday I
      • 09:00
        Welcome 15m
        Speaker: Andreas Søgaard
      • 09:15
        GraphNeT overview 30m

        Introduction to the GraphNet package: Vision, status, achievements, plans, roadmap

        Speaker: Andreas Søgaard
      • 09:45
        Representation learning using Graph Neural Networks 45m

        Learning compact and continuous representation spaces for discrete and non-Euclidean objects such as graphs can be useful to study their properties and perform meaningful vector manipulations that can translate back into the data space. In this session, we will look into graph neural network based regularised auto-encoders that can learn embedding spaces, that can also be conditioned on additional physical properties.

        Speaker: Raghavendra Selvan (DIKU)
    • 10:30 11:00
      Coffee 30m
    • 11:00 12:30
      Presentations: Tuesday II
    • 12:30 13:30
      Lunch 1h
    • 13:30 17:00
    • 17:00 18:00
      Presentations: Poster session
      • 17:00
        Poster - Improving the direction and energy estimation of low-energy events in ANTARES with machine learning 5m
        Speaker: Juan García Méndez (Universitat Politècnica de València)
      • 17:05
        Poster - Impact of muon deflections on angular muon reconstructions 5m
        Speaker: Pascal Gutjahr
      • 17:10
        Poster - Detection and reconstruction of GeV neutrinos with IceCube 5m
        Speaker: Karlijn Kruiswijk
      • 17:15
        Poster - Improving the sensitivity of KM3NeT in the MeV-GeV range using a GCN 5m
        Speaker: Jonathan Mauro
      • 17:20
        Poster - Ideas to measure the prompt component of the atmospheric muon flux 5m
        Speaker: Ludwig Neste (TU Dortmund University)
    • 19:00 22:00
      Social: Workshop dinner
    • 08:30 09:00
      Arrival and coffee 30m
    • 09:00 10:30
      Presentations: Wednesday I
      • 09:00
        GraphNeT results 45m

        Overview of physics analyses and studies that have been carried out in the context of GraphNeT, their impact on physics, and possible future directions.

        Speaker: Rasmus Ørsøe
      • 09:45
        PyG 2.0 45m
        Speaker: Matthias Fey (PyG and kumo.ai)
    • 10:30 11:00
      Coffee 30m
    • 11:00 12:30
      Presentations: Wednesday II
    • 12:30 13:30
      Lunch and workshop photo 1h
    • 13:30 17:00
    • 17:00 22:00
      Social: Own time
    • 08:30 09:00
      Check-out and coffee 30m
    • 09:00 10:00
      Hackathons: Presentation of results
    • 10:00 10:30
      Presentations: Friday I
      • 10:00
        Roadmap towards v2.0 – mini-workshop introduction 30m
        Speaker: Andreas Søgaard
    • 10:30 11:00
      Coffee 30m
    • 11:00 12:30
      Presentations: Friday II
      • 11:00
        Roadmap towards v2.0 – mini-workshop 1h 10m
        Speaker: Andreas Søgaard
      • 12:10
        Summary, next workshop, and goodbye 20m
        Speaker: Andreas Søgaard
    • 12:30 13:30
      Lunch 1h
    • 13:30 16:30
      Social: Optional excursion