Hubert Baniecki is a PhD student at the University of Warsaw and was a visiting researcher in project C03 of TRR 318 at the LMU Munich in March. His research focuses on the explainability of AI systems—specifically on efficient methods for explaining machine decisions and their robustness. In the interview, he shares what he researched with the TRR members, what made the collaboration special for him, and the impressions he took away.
TRR 318: What did you work on during your time at LMU Munich?
Hubert Baniecki: “During my stay in March, I collaborated with Maximilian Muschalik, Fabian Fumagalli, and Professor Eyke Hüllermeier on a joint research project. We investigated the possibility of explaining vision-language models with Shapley interactions, which is a direct continuation of our recently published work. One key result of the stay was scaling the {shapiq} framework to handle deep learning problems more efficiently.”
What goal are you pursuing with that?
“The goal of the collaboration is to develop a new way of interpreting decisions made by complex AI systems that jointly process image and text inputs. Think of an image of a bird and a caption describing the feathers and color of the bird. We also aim to include an implementation of the new algorithm within the open-source {shapiq} software package.”
What made the collaboration special for you?
“The collaboration was incredibly productive—we essentially completed two months of work in just one, thanks to the intensive pace and close coordination with the team. It wouldn’t have been possible without the exceptional game-theoretic expertise of project member Fabian Fumagalli. Also, the discussions with Professor Hüllermeier were very insightful and helped shape and expand the direction of the project.”
What do you take away from your stay?
“I really enjoyed bonding with colleagues, exchanging experiences, and being immersed in a new research environment for a month. I also had the opportunity to attend the group seminar and present my ICLR spotlight paper. This was a great chance to share my findings to get feedback and discuss natural future work.”
Further Information:
- TRR 318 Research Training Group
- TRR 318 Project C03
- Information on the Visiting Fellowship