Carsten T Lüth

I’m finishing my PhD at Heidelberg University, the Division of Medical Image Computing at DKFZ & Helmholtz Imaging (defense: April 2026), supervised by Prof. Klaus Maier-Hein and Dr. Paul Jaeger (Google DeepMind).

The throughline of my research is a simple question:

Are we actually measuring what we think we’re measuring?

This has taken me through Active Learning, Uncertainty Estimation, and Selective Classification — always asking whether the evaluation practices the field relies on are telling us the truth. The answer is often no, and I try to do something about it. This work has been recognized with 2 ICLR Oral presentations, a NeurIPS Spotlight, and a MICCAI STAR Award (~700 citations total).

The most concrete expression of this is nnActive — the largest benchmark for Active Learning in 3D biomedical segmentation (~150k A100 GPU hours), built on top of nnU-Net, the segmentation framework I help maintain (8k+ stars, 1,000+ active users worldwide).

I also organize heidelberg.ai, a community of 3,000+ AI practitioners in Heidelberg. We’ve hosted talks from the teams behind Stable Diffusion, AlphaFold, and Aleph Alpha, and I’m proud of the space we’ve built for genuine scientific exchange.

On a personal note: I read widely (fiction, statistics, psychology, philosophie — in no particular order), and outside of work I surf, play piano, cook, and spend time with friends.