Best Paper Award
Congratulations to Dr. Clemens Lutz, Christina Dionysio, Louis Le Page, Ben-Oliver Hosak, Elias von Brunn, Zeynep von Brunn, Yuka Nakamoto, Muaid Mughrabi, who were each honored with an award for their scientific work at the BTW conference, which took place in Bamberg March 3 — 7, 2025.
Christina Dionysio, a PhD student in the research group of Prof. Matthias Böhm, together with Louis Le Page, was awarded the Best Paper Award for the publication "Scalable Computation of Shapley Additive Explanations" (DOI: 10.18420/BTW2025-16).
“This work addresses the computational challenges of SHAP, which is a widely used model-agnostic method for explaining local and global machine learning model predictions. We optimize Antithetic Permutation Sampling through vectorized and parallelized operations on Apache SystemDS. This approach leads to a significant improvement in runtime, achieving up to 14 times faster performance on single-node operations and up to 35 times faster performance on distributed operations, without compromising explanation accuracy. By leveraging these optimizations our approach enhances the efficiency and applicability of SHAP for large datasets and complex models”, explains Christina.
Data Science Challenge
Ben-Oliver Hosak, Elias von Brunn, Zeynep von Brunn, Yuka Nakamoto, Muaid Mughrabi, all students from the team of Prof. Ziawasch Abedjan X secured second place in the Data Science Challenge and brought it back to Berlin. "This year’s challenge at BTW was to predict hourly energy prices for an entire day (24 predictions, one for each hour). Our solution had several key contributions. Since energy prices depend on supply, demand, and energy sources, we started by gathering data from various relevant sources, including energy mix, weather data, and historical prices. We then cleaned the datasets and performed feature engineering. To determine whether a more complex model was worth the investment, we experimented with three models of varying computational complexity. Another important contribution was our test benchmark, which simulated the challenge submission by requiring 24 consecutive predictions. We tested different scenarios with missing features and submitted the model that demonstrated the most robust behavior when some data was unavailable at submission time", explains Muaid Mughrabi.
Best Dissertation Award
BIFOLD alumni Dr. Clemens Lutz received the Best Dissertation Award 2024 for his outstanding doctoral research in the field of database systems. This award was conferred on him by the Fachgruppe Datenbanken and Informationsssyteme (DBIS-Dissertationspreis) of the Gesellschaft für Informatik as an outstanding recognition for his work (please also read here).