Publications
Novel Ion Suppression-Based MS Approach (IS-MS) for Straightforward, Robust, and Rapid Identification and/or Differentiation of Liquid Samples: Wine as a Case Study
Analytical Chemistry, June 2026, 98 (24), 18010–18018
Joint work with Maria Bikaki, Lu Hostettler, Matthias Kleiner, Götz Schlotterbeck, and Markus Ehrat.
We developed a method for authenticating liquid samples using the distinctive patterns that ion suppression produces in mass spectrometry. Using wine as a case study, we combined these fingerprints with statistical analysis to distinguish wines by vintage, origin, and producer, with measurements taking about one minute per sample.
Read the Paper, Write the Code: Agentic Reproduction of Social-Science Results
arXiv preprint, April 2026
Joint work with Benjamin Kohler, Johanna Einsiedler, Alexander Hoyle, and Elliott Ash.
We investigate whether LLM agents can reproduce social-science findings from methods descriptions and original data. Our system extracts the methods, has agents write analysis code without access to the original code or results, and compares their outputs with published findings. Across 48 papers, agents recover many results, while errors reveal both limitations of the agents and missing details in the papers' methods.
Beyond Pairwise Correlations: Higher-Order Redundancies in Self-Supervised Representation Learning
arXiv preprint, December 2024
Joint work with Béni Egressy, Frederik Benzing, Matthias Otth, and Roger Wattenhofer.
We investigated redundancy patterns in self-supervised learning, going beyond simple correlations to measure and reduce more complex dependencies. Our work introduces SSLPM, a novel approach where neural networks compete to minimize predictable patterns, providing new insights into the role of redundancies in self supervised learning.
Synthetic Cancer - Augmenting Worms with LLMs
Submission to the Swiss AI Safety Prize, February 2024
Joint work with Benjamin Zimmerman.
🥇🎉 We are thrilled to announce that we have won the first prize and received 7'500 CHF!
Our submission presents a novel malware approach that enhances worms with Large Language Models (LLMs). The proposed metamorphic malware utilizes LLMs for code rewriting and targeted spreading through social engineering.
- Contest Submission Paper
- Coverage by "New Scientist".
- Swiss National Radio SRF1 "Echo der Zeit" Radio Coverage on the Prize and our Research.
- Coverage on Medium.
Towards a Deeper Understanding of Semantic Comprehension in Language Models Paired with Semantic Graphs
Joint work with Matthias Kleiner and Ahmet Özüdogru.
We extend previous work by (Wu et al., 2021) who proposed infusing additional semantic information in the form of DM graphs into graph neural networks stacked on top of language models into models for language tasks. We present a detailed ablation study on semantic understanding in these models and extend them using AMR graphs.
ETH Materials
This is a collection of some materials I have written while studying at ETH Zürich. Since they might be useful for others too, I have decided to make them available on the internet.
More links will follow in the future.
More Material
If you are looking for more I can wholeheartedly recommend Sven Pfiffner's website.
Analysis 1 & 2
For the ETH Ananlysis lectures I have written two summaries which I have used in the exams:
Licensed under the permissive MIT License. Your help is greatly appreciated! Feel free to send me an email or to open an issue when you find mistakes.
Linear Algebra
A handwritten Linear Algebra summary written for the final exam. The summary was written in fall of 2018.
Probability and Statistics
For spring's Probability and Statistics lecture given by Prof. Teichmann I have written a short summary with key methods and theorems.
The summary is in German as the class was taught in German!
Licensed under the permissive Attribution-ShareAlike 4.0 International. Your help is greatly appreciated! Feel free to send me an email or to open an issue when you find mistakes.
Numerical Methods for CSE
A summary which I have taken to the exam. It mainly consists of Eigen documentation and concepts that could not easily be found on the provided lecture notes.
The summary might have some rough edges.
Licensed under the permissive Creative Commons Attribution-NonCommercial 4.0 International License. Your help is greatly appreciated; Feel free to send me an email or to open an issue when you find mistakes.
Promela Cheatsheet
A small summary with the main language constructs of the promela language used by the spin model checker. Written in spring of 2020 for the class "Functional Programming and Formal Methods" at ETH Zürich.
Licensed under the permissive MIT License. Your help is greatly appreciated; Feel free to send me an email or to open an issue when you find mistakes.
SQL Cheatsheet
A quite comprehensive SQL cheatsheet covering all SQL constructs covered in class. Written in Spring of 2020.
Licensed under the permissive MIT License. Your help is greatly appreciated; Feel free to send me an email or to open an issue when you find mistakes.
Algorithms Collection for Freshmen Algorithms Classes at ETH
During the first year of my CS studies I have kept a repository where I have implemented a majority of the algorithms shown in class.
The repository definitely has some rough edges as many algorithms were implemented under time constraints, but all in all it should contain a wide variety of algorithms.
ETH Introduction to Machine Learning Cheatsheet 2021
Exam summary for the course Introduction to Machine Learning (2021) at ETH Zürich. The summary is an adapted and slightly updated version of the outstanding summary eth-cs-student-summaries / Introduction-to-Machine-Learning summary. Please check out the original version!