Juelich Challenges - FZJ Hackathon
Based on existing experiences in creating, hosting and advertising plant science data challenges, a group of JULAIN researchers has been formed to publish further data challenges and to invent 'Jülich Challenges’ as a label for our unique scientific data and according research problems.
‘Jülich Challenges’ target ML researchers, or traditional image analysts, looking for interesting data to work on, that allow quick results for fast publications. As benefits for FZJ we expect to attract ML and AI specialists to our unique data sets, scientific questions and therewith to provide solutions for Jülich-specific research questions. This comes with an increased visibility in the fields of ML and AI - across domains and within the ML communities.
Scope
With help of Jülich's Helmholtz AI consultant team the Jülich Challenges platform was implemented to enable the data science challenges to run: a website that acts as a central interface, describes the challenges in detail, offers data and scripts for download, and evaluates and suitably presents the results. The first challenges have been integrated, with more to follow.
Before we officially release platform and challenges and advertise under the label Jülich Challenges we will run a FZJ internal hackathon - put it to the acid test and come up with baseline solutions.
In the frame of the hackathon a short introduction on running ML models on JSC's HPC systems will be offered. For the time of the hackathon we will provide compute resources through HAICORE to our participants. The Helmholtz AI Consultant team will guide you through the required steps on the supercomputer. Taking part can be a great first step in bringing your favourite research topic to the fastest computer in Europe.
All Machine Learning interested colleagues, especially MSc and PhD students, are invited to attend.
If you're curious, please have a look on already implemented challenges at the Jülich Challenges Platform. By the time of the hackathon probably two more challenges will be available: one about the prediction of sex from whole Brain Connectome matrices from the 1000 Brains Study, the other about predicting live time of thermal barrier coatings from 2D images showing the microstructure of the ceramic thermal barrier coating layers.
 Institute of Bio- and Geosciences, Plant Sciences (IBG-2) |
 Jülich Supercomputing Centre |
|
 Institute of Bio- and Geosciences, Agrosphere (IBG-3) Root-Soil Interaction group |
 Institute of Neuroscience and Medicine, Brain and Behaviour (INM-7) |
 Institute of Energy and Climate Research, Materials Synthesis and Processing (IEK-1) |
Schedule
The hackathon is scheduled for 4 days as a fully virtual event. The first day is dedicated to the introduction of the platform and available challenges as well as aforementioned introduction into Jülich's Supercomputers. You will form teams for collaborative coding, ideally mixed between different working groups or institutes, and pick a challenge you'd like to work on. Of course you can join as a team as well. The remaining time will be used for working on the challenges. At the end of the week we will hopefully have models that solve our challenges up to a certain performance and therewith proof their feasibility but also its challenging character.
Contact persons for the data sets as well as mentors to assist with ML and HPC related questions will be around to support you.
We are pleased to be able to offer tutorials and lectures on each day. A big thanks to Stefan Kesselheim, Andreas Herten, and Claudia Comito.
Tuesday
- 9:00am Join us at GatherTown, explore the space
- 9:10am Welcome session at the auditorium (Zoom) with pitches to each challenge
- 9:45am Registration to Jülich Challenges Platform
- 10:15am Detailed discussion about individual challenges with our data providers in breakout rooms, getting to know interested researchers and form teams
- 11-12am Tutorial: Machine Learning with HPC, Stefan Kesselheim (JSC)
- Afternoon: Happy Coding
Wednesday
- 9am Daily Stand-up meeting at the auditorium
- 11am Tutorial: Data-parallel training, Stefan Kesselheim (JSC)
- Coding
Thursday
- 9am Daily Stand-up meeting
- 11am Talk: GPU Computing 101, Andreas Herten (JSC)
- Coding
Friday
- 9am Daily Stand-up meeting
- 11am Talk: HeAT: a distributed tensor framework for high-performance data analytics, Claudia Comito (JSC)
- 4pm: Closing
Links to the gather space and zoom will send out the day before the hackathon. Please keep an eye on your emails.
The event is free of charge. Coffee, cookies, and donuts you have to bring your own unfortunately. Please register to stay up to date with related information by Monday 15 March, 2021 at the latest.
Sponsors
Gather is building virtual spaces for people to live and interact more effectively online, while having more fun. We combine video-calling with a 2D map, letting you walk around and talk to the other people right next to you. Our creation tools also let you make your own custom, interactive spaces!
