Home
I am a PhD student at the Max Planck Institute for Intelligent Systems, ELLIS Institute Tübingen, and the University of Tübingen, affiliated with the Tübingen AI Center, and funded by the Cluster of Excellence Machine Learning. I am an organizer of FridayTalks@Tübingen, a bi-weekly research talk series for the Tübingen AI ecosystem.
I am fortunate to be supervised by Jonas Geiping and I was previously supervised by Robert Bamler. I also spent four months as a research intern in Roger Grosse’s group at the Vector Institute. Currently, I am most interested in large language models, in particular post-training, test-time scaling, sampling, and (probabilistic) inference.
I am entering the industry job market at the end of this year. Please reach out if you know of relevant opportunities in large language models, especially around post-training, test-time scaling, sampling, and inference.
News
- July 2026: I will be attending ICML — reach out to discuss anything related to LLMs, inference, or an industry position!
- June 2026: We released a preprint on sequence probability and correctness in LLMs; the corresponding workshop paper got an oral presentation at ICML’s SPIGM workshop!
- April 2026: We released an extended version of our workshop paper generalizing its sampling method.
- March 2026: I contributed to a new workshop paper that introduces ε-Leaf Enumeration, a deterministic tree-search alternative to self-consistency that avoids repeated reasoning prefixes and shows performance improvements on math, coding, and general reasoning tasks. The paper will be presented at the Workshop on Latent & Implicit Thinking at ICLR!
- July 2025: A new workshop paper that I co-authored will be presented at RLC PRL and EWRL18! In the paper, we study the use of LLMs for traditional AI planning. Check it out if you’re interested in whether, and how, LLMs can tackle AI planning tasks!
- June 2025: In a new preprint I co-authored, we show that large language models struggle with sampling from Bernoulli distributions and introduce a rejection-sampling-based method that effectively reduces this bias. It’s a nice read if you are interested in (rejection) sampling and/or LLMs!
- January 2025: We finally managed to find the time to publish an extension of my Master’s thesis as a paper (preprint) at a chemistry journal.
- January 2025: We relaunched FridayTalks@Tübingen, a bi-weekly AI research seminar for the Tübingen AI ecosystem.
- December 2024: Our paper, A Note on Generalization in Variational Autoencoders: How Effective Is Synthetic Data and Overparameterization? has been accepted at TMLR following a smooth and constructive review process! Check it out if you (still) like VAEs!
- September 2024: I am excited to be joining the group of Roger Grosse for the Vector research internship in February 2025! Please reach out if you are based in the Toronto area!
- June 2024: I will present our paper at ICML 2024 and AABI 2024 in person. Chat with me in Vienna! I will also present the project at the Approximate Inference in Theory and Practice workshop in Paris!
- May 2024: Our paper Differentiable Annealed Importance Sampling Minimizes The Symmetrized Kullback-Leibler Divergence Between Initial and Target Distribution has been accepted at ICML 2024!
- November 2023: I am visiting NeurIPS 2023 in person. Talk to me in New Orleans!
- October 2023: Our paper The SVHN Dataset Is Deceptive for Probabilistic Generative Models Due to a Distribution Mismatch has been accepted at the NeurIPS 2023 Workshop on Distribution Shifts! Check out the project page!
- May 2023: I am visiting ICLR 2023 in person. Come and talk to me in Kigali, Rwanda!
- April 2023: Our paper Resampling Gradients Vanish in Differentiable Sequential Monte Carlo Samplers has been accepted as a TinyPaper at ICLR 2023!
- July 2022: I am starting a PhD in the group of Robert Bamler!