Review
Challenging topics for an inquisitive group: the first ErUM Data Hub Active Training Course "Advanced Deep Learning"
With the Active Training Course "Advanced Deep Learning", the ErUM Data Hub closes the event year 2022. We look back on a successful event.
From November 28th to December 1st, 2022, a total of 40 senior scientists, postdocs, PhD students, and undergraduates gathered at Landhaus Nordhelle in Meinerzhagen to explore advanced Deep Learning concepts together. With participants from 6 ErUM communities (KAT, KET, KfB, KFS, KHuK, and RDS), the group also brought a variety of professional perspectives to this event.
The 4 days in Landhaus Nordhelle started with a first dinner on Monday. Afterwards, the ErUM Data Hub team (Prof. Dr. Martin Erdmann, Angela Warkentin and Peter Fackeldey) gave a short welcoming to the participants and handed over to Prof. Dr. Tobias Golling (University of Geneva), who started with a late-night lecture on the subject of Graph Neural Networks .
After an invigorating breakfast, the participants were able to try out the input from the first lecture in a hands-on session on Tuesday morning. During the coffee break that followed, insights were exchanged and energy was recharged for the next block of topics. Afterwards, Niklas Langner (RWTH Aachen University) took the floor and introduced the topic of Transformers . This content was also supplemented by a practical exercise after lunch.
The participants used the remaining Tuesday afternoon for a round of disc golf in the fresh November air and to get into a competitive mood at the in-house bowling alley.
Wednesday was led by Dr. Claudius Krause (University of Heidelberg) with a lecture and hands-on session on the topic of Normalizing Flows . Prof. Dr. Michael Krämer (RWTH Aachen University) concluded the theoretical part with a lecture on Autoencoders. Afterwards, the participants were given the opportunity to put the newly acquired knowledge of the last few days to the test.
The ErUM-Data-Hub Milky-Way Challenge
The 5 drawn groups quickly got together and familiarized themselves with the task designed by Peter Fackeldey. The first approaches to solving the problem were discussed and only reluctantly did the teams allow themselves to be interrupted by dinner. Afterwards, the participants continued to devote themselves to the challenge with great dedication. The light in the group room burned late into the night.
Thursday morning continued to be dedicated to the Challenge. The groups put the finishing touches to their solutions and each prepared a 10-minute presentation on their approach and result. A last fortifying coffee was served and the 5 presentations were held. The judges Dr. Claudius Krause and Peter Fackeldey did not find the evaluation easy. At the award ceremony they emphasized the impressive performance of all teams. The winning team was happy to receive a small gift from the ErUM Data Hub.
After concluding words from the ErUM Data Hub team and a final get-together over lunch, participants said their goodbyes and headed home.
We say "thank you"
We would like to thank all speakers and the tutors Moritz Scham, Nathan Prouvost and Sascha Daniel Diefenbacher for their valuable input and active support, as well as all participants who contributed to the success of the first Active Training Course "Advanced Deep Learning" through their diligent cooperation. We would also like to thank the BMBF for their financial support.
An eventful and successful first year lies behind the ErUM Data Hub. We are happy about the positive feedback on our events and look forward to an exciting year 2023 - with a larger team and new workshops, schools and training courses.
