Machine Learning and Deep Learning in Science & Physical Systems

Over two joint parallel sessions at the upcoming D3A meeting, we'll provide
a sequence of short talks, lightning introductions, practical discussion
and hands-on demos of topics in these areas.


Introduction: Physics-informed neural networks

Presented by Prof. Allan P. Engsig-Karup (DTU), a 30-minute introduction to the connections between machine-learning and physics.


Applications: Machine learning and data science in practice

Nikolas Borrell-Jensen (DTU) presents recent work on sound field predictions with PINNs; and Benedikt Sommer (Maersk/Novo Nordisk) presents work from Maersk on forecasting and decision-making for empty container repositioning.

Lightning Introductions

We warmly encourage each participant to present a one-minute, one-slide personal introduction to their interests or research. We will solicit material in advance, and identify common themes for discussion in the following session.

Geometric Deep Learning for Science

The Geometric Deep Learning for Science workshop provides hands-on demonstrations and discussions between experts and newcomers to geometric deep learning (GDL). We will uncover common challenges, tools, and methods across Danish research groups in physics, chemistry and biology.


What are the biggest challenges of your domain (e.g. high energy physics, computational biology, materials chemistry) and how could these problems be presented in terms of geometric deep learning?

A discussion leader will connect the dots between talks and those in the room, initiate discussions amongst attendees, and suggest directions for research and collaboration. We emphasise especially that we can take advantage of this conference being a ‘for-Denmark’ meeting, to encourage a high degree of openness about challenges, and willingness to learn.

Hands-on demonstrations

A set of three concrete examples presented in live coding sessions, showing the details and potential to apply to diverse datasets. The goal is to demystify the machinery behind geometric deep learning: graph neural networks, transformers, symmetry-constrained learning.

By presenting these tools in an accessible way, with open-source libraries behind each demo, and further material for attendees to follow after the session, we hope to spur researchers new to GDL to apply these tools, and to encourage expert researchers to see how the tools are applied in other domains.

Register to participate!

We're excited to welcome an active group of participants with diverse backgrounds, interests and goals for these sessions. The only requirement for participating is a strong enthusiasm for learning and contributing to making the meeting a success.

Please fill in your details below to sign up for communication about these sessions. We strongly encourage contribution of a single-slide for the lightning introductions and will be in touch by email during January to finalise these.

  • Organisers

  • Geometric Deep Learning

    Daniel Murnane, NBI
    Troels Petersen, NBI
    Inar Timiryasov, NBI
  • Machine Learning in Physics

    Berian James, DTU & Pioneer Centre
    Julija Tastu, A.P. Møller-Maersk
    Stefan Pollok, DTU Energy