Spotlight Talk at AI For Tabular Data Workshop @ EurIPS 2025 🇩🇰

1 minute read

Published: December 06, 2025

Tusind tak København! 🇩🇰

I am incredibly honored to have presented our work, “Exploring Multi-Table Retrieval Through Iterative Search,” as a Spotlight Talk followed by a poster session at the AI for Tabular Data Workshop at EurIPS 2025 in Copenhagen on December 6th.

It was a fantastic experience to engage in insightful discussions with fellow researchers and industry professionals. We dove deep into real-world challenges in multi-table retrieval and how our future work can help solve meaningful problems in this space.

In addition to presenting, I also had the privilege of serving as a Program Committee (PC) member for the workshop. Being on the other side of the review process was a rewarding experience that gave me a deeper appreciation for the effort that goes into curating a high-quality technical program.

About The Work

Multi-table retrieval focuses on gathering and composing information from multiple tables to answer complex natural language questions. Our work introduces an iterative and explorative strategy which brings several key benefits:

🚀 Scalability: It enables efficient heuristics that avoid costly exact optimization. As shown across 5 NL2SQL benchmarks, our method is 4 to 400 times faster while maintaining competitive performance.
👀 Interpretability: Each step acts as a visible reasoning checkpoint, making the retrieval process transparent.
🧩 Flexibility: It opens the door for extensibility and potential human-in-the-loop guidance.

Acknowledgements

This work wouldn’t be possible without my supervisors and co-authors from the Database team at LIP6 - Laboratoire d’informatique: Bernd Amann, Rafael Angarita, and Hubert Naacke.

A sincere thank you to the workshop organizers — Madelon Hulsebos, Effy Li, Lennart Purucker, Peter Baile Chen, and Frank Hutter — for bringing this community together. The spotlight talks and keynotes were on fire, especially the exciting updates regarding TabPFN!