Home
About me
I am a researcher working on optimization for machine learning. Currently, I am a member of technical staff in pretraining at Cohere. Previously, I obtained my PhD at Technical University of Munich (advisor: Michael Ulbrich) and was a postdoc at Inria Paris (advisors: Francis Bach, Umut Simsekli, and Adrien Taylor).
Research Overview. A common theme of my past work is to understand, simplify, and ultimately improve training recipes for ML models. Selection of past work:
- Understanding the behaviour of learning-rate schedules for LLM training [ICML 2025].
- What are good optimizers for training diffusion models? [EurIPS 2025]
- Training algorithms that need less tuning (MoMo [ICML 2024], proximal Polyak step size [TMLR 2023]).
Contact
You can write me at firsstname[dot]lastname[at]tum[dot]de or find me on Strava.