March 19, 2023 - ETP-Team wins 1st AKPIK Datathon

In the context of the 83th Spring Meeting of the German Physical Society the 1st Datathon of the Working Group on Physics, Modern IT and Artificial Intelligence (AKPIK) was held. Together with 8 other teams, Greta Heine, Cedric Verstege, Moritz Bauer, Florian Wemmer and Jonas Eppelt faced the challenge.
A modified Fashion-MNIST training dataset and four hours time were given. The task was to select the best 10.000 datapoints in the training dataset to train a given neural network and achieve the highest accuracy on a test dataset.
The ETPs team followed multiple paths to solve this task and tried Autoencoders, CNNs, and manual sorting. In the end a simple Lasso-Regression proved to be the best solution of all 46 handed in solutions.
Congratulations also to the Team “54th Element” for winning the most creative solution price. All results can be found at https://cloud.e5.physik.tu-dortmund.de/akpik23.
Thanks to the organisers from AKPIK and the sponsor d-fine. We are already excited for the 2nd Datathon.