Hi, I am a PhD candidate in Machine Learning at ETH Zurich and the Max Planck Institute for Intelligent Systems. I am advised by Andreas Krause and Bernhard Schölkopf. Currently, I am visiting Google's Paradigms of Intelligence team as a student researcher.
My research focuses on post-training language models to improve their reasoning capabilities and alignment, in particular through reinforcement learning and self-distillation. I am part of the team that developed Apertus, a fully open family of language models at 8B and 70B scales.
Beyond this, I am also interested in causal inference and causal reasoning.
I hold a Master's degree from ETH Zurich. For my Master's thesis, I worked with Charlotte Bunne and David Alvarez-Melis at Harvard on optimal transport for modeling single-cell dynamics. Before my PhD, I was a Data Scientist at QuantCo and interned at IBM.
Reach out if you want to chat!
Email: frederike.luebeck@inf.ethz.ch
News
-
September 2026
I joined Google’s Paradigms of Intelligence team as a student researcher.
-
September 2026
New blogpost and arxiv paper on self-distillation with attractive and repulsive teachers, and how combining the two isolates the correctness signal and cancels unintended behavioral shifts.
-
August 2026
The Apertus 1.5 models are available on Hugging Face, now with multimodal input and reasoning support. Tech report coming soon.
-
May 2026
Happy to see that our self-distillation method was used by Cursor to train their new model Composer 2.5
-
January 2026
Our paper on Reinforcement Learning via Self-Distillation is out
-
September 2025
Apertus is on Hugging Face, along with the tech report. Building an LLM from scratch was a huge team effort, and it was a lot of fun to do it with such an incredible team!
Publications & Projects
-
On Repulsive and Attractive Teachers: Separating Correctness from Behavior in Self-Distillation
Anton Baumann, Akmal Ashirmatov, Leo Schmidt-Traub, Frederike Lübeck, Jonas Hübotter, Thomas Kleine Buening, Andreas Krause
Preprint, 2026
-
Reinforcement Learning via Self-Distillation
Jonas Hübotter, Frederike Lübeck, Lejs Behric, Anton Baumann, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, Andreas Krause
ICML, 2026
Best Paper Award at ICLR 2026 Workshop on Test-Time Updates
Model
Master Thesis Projects
I am happy to supervise Master students. You can find current available projects on the LAS Group Student Projects page. If you're interested, please reach out with your research interests, along with your up-to-date CV and academic transcript.