A central activity at the workshop will be two half-day "hackathons" where attendees can collaborate hands-on on implementing GNNs in their physics analysis, experimenting with new analysis approaches, developing new reconstruction techniques, or similar — all using the GraphNeT framework.
We encourage participants to contact the organiser (Andreas Søgaard) with ideas for hackathon topics they would be interested in working on. For instance, you could work on:
- getting GraphNeT installed and training your first GNN model. The organisers will prepare some standard, experiment-independent data based on the recent IceCube Kaggle challenge. This will allow anyone to get up and running quickly.
- applying GraphNeT to data from your own experiment (that is not IceCube). In this case you should, in advance, consider which data could be used to test this out, and ideally bring it along for the hackathon. BYOD — bring your own data.
- improving performance on a certain reconstruction task by adding new GNN operations, layers, etc. You can do this on the standard Kaggle or the BYO data.
- etc.
At the workshop, we will present a collection of potential topics (including ones that might have come up during the first half-day) and invite attendees to pick one to focus on, with a goal to achieve concrete results by the end of the workshop. The GraphNeT developers will be around to help you get setup, running, and and extending the framework.
On Thursday morning, teams are encouraged to present their hackathon results.
