Martin Wertich
I am an MSc student in Computer Science at ETH Zurich working on machine learning theory, with a focus on learning-theoretic questions, high-dimensional statistics, and the mathematical structure of modern ML systems.
My goal is to better understand why advanced learning methods generalize, how they behave under uncertainty, and which mathematical principles make them expressive and robust.
Current Work
From February 2025 - April 2026, I worked for the ETH AI Center under the supervision of Barna Pásztor, Ido Hakimi, and Prof. Andreas Krause in the Learning and Adaptive Systems Group. There, I contributed to ActiveUltraFeedback, an RLHF pipeline for large language models based on active learning. The resulting paper was accepted to ICML 2026 and is available on arXiv.
I am currently working on a semester project under Julia Kostin and Prof. Fanny Yang on theoretical guarantees for compositional generalization under noise in additive energy-based settings.
Research Interests
My main research interest is machine learning theory, especially problems that combine mathematical rigor with clear implications for modern learning systems. I am particularly drawn to learning theory, random matrix/graph phenomena, and the statistical structure underlying high-dimensional models.
What I find most compelling about theoretical machine learning is that it forces us to move beyond intuition alone. Many of the most successful methods in modern AI work far better than our theory currently explains. Bridging that gap, even incrementally, is what motivates me.
Background
I completed my bachelor’s degree at Johannes Gutenberg University Mainz, where I became increasingly interested in the mathematical side of computer science. My bachelor’s thesis focused on the mathematical structure of attention layers in Transformer models and introduced me more deeply to the theoretical questions that now shape my work.
At ETH Zurich, I have complemented this theoretical focus with teaching and research experience. I worked as a teaching assistant for Stochastics and Machine Learning and as an assistant for Introduction to Machine Learning, where I contributed to improvements to Ethel.
Before and alongside academia, I also gained industry experience in machine learning and software engineering. At Schott, I worked on explainable AI for large time-series data streams, and at Envision Entertainment, I contributed as a software developer to the strategy game Pioneers of Pagonia.
Outside Research
Outside academic work, I enjoy hiking, climbing, skiing, and cross-country skiing, especially in the Swiss Alps. I also like board and card games, which are an excellent excuse to spend time with friends and colleagues.