Skip to content
Technical University of Munich

Yan Scholten

Technical University of Munich
Department of Computer Science - I26
Boltzmannstr. 3
85748 Garching b. München
Germany

Room: 00.11.065
E-Mail: y.scholten [at] tum.de
Website: yascho.github.io
 

Research Interests

My research centers on trustworthy AI, with a focus on developing methods that make machine learning more safe, reliable, and aligned with human values. I tackle core challenges in AI, such as machine unlearning, alignment, adversarial robustness, robustness certification, and conformal prediction. My recent work advances the capabilities and reliability of large language models (LLMs).
 

Selected Publications

Full list on Google Scholar

  • Model Collapse Is Not a Bug but a Feature in Machine Unlearning for LLMs
    Yan Scholten, Sophie Xhonneux, Leo Schwinn*, and Stephan Günnemann*
    [Preprint | Project page | Code | Blogpost ], 2025
     
  • Sampling-aware Adversarial Attacks Against Large Language Models
    Tim Beyer, Yan Scholten, Leo Schwinn*, and Stephan Günnemann*
    [Preprint | Project page], 2025
     
  • Provably Reliable Conformal Prediction Sets in the Presence of Data Poisoning
    Yan Scholten, Stephan Günnemann
    International Conference on Learning Representations, ICLR 2025 (Spotlight)
    [PDF | Project page | Code]
     
  • A Probabilistic Perspective on Unlearning and Alignment for Large Language Models
    Yan Scholten, Stephan Günnemann, Leo Schwinn
    International Conference on Learning Representations, ICLR 2025 (Oral)
    [PDF | Project page | LLM framework (Code) | Confidence bounds (Code)]
     
  • Adversarial Alignment for LLMs Requires Simpler, Reproducible, and More Measurable Objectives
    Leo Schwinn, Yan Scholten, Tom Wollschläger, Sophie Xhonneux, Stephen Casper,
    Stephan Günnemann, Gauthier Gidel
    [Preprint], 2025.
     
  • Hierarchical Randomized Smoothing
    Yan Scholten, Jan Schuchardt, Aleksandar Bojchevski, Stephan Günnemann
    Advances in Neural Information Processing Systems (NeurIPS), 2023
    [PDF | Project page | Code]
     
  • (Provable) Adversarial Robustness for Group Equivariant Tasks:
    Graphs, Point Clouds, Molecules, and More
    Jan Schuchardt, Yan Scholten, Stephan Günnemann
    Advances in Neural Information Processing Systems (NeurIPS), 2023
    [PDF | Project page | Code]
     
  • Randomized Message-Interception Smoothing: Gray-box Certificates for Graph Neural Networks
    Yan Scholten, Jan Schuchardt, Simon Geisler, Aleksandar Bojchevski, Stephan Günnemann
    Advances in Neural Information Processing Systems (NeurIPS), 2022
    [PDF | Project page | Code ]
     

Research Experience

  • May 2025 - Jul 2025: Research visit, Carnegie Mellon University, USA
  • May 2018 - Jul 2018: Undergraduate research visit, University of Western Ontario, Canada
  • Oct 2017 - Apr 2018: Undergraduate research assistant, Paderborn University, Germany
     

Education

  • 2022-now: PhD student in Computer Science, Technical University of Munich
  • 2019-2022: M.Sc. Informatics - Technical University of Munich (passed with high distinction)
  • 2015-2019: B.Sc. Computer Science (Math Minor) - Paderborn University (passed with distinction)
     

Theses

  • Master's thesis (2022): Interception Smoothing: Gray-box Certificates for Graph Neural Networks
  • Bachelor's thesis (2019): Towards a Large-Scale Causality Graph
     

Academic Honors and Awards

  • 2025: Selected as Top Reviewer at NeurIPS 2025
  • 2023: Admission to the Konrad Zuse School of Excellence in Reliable AI
  • 2019: Deutschlandstipendium awarded by the Technical University of Munich
  • 2018: RISE worldwide scholarship awarded by DAAD
  • 2018: Deutschlandstipendium awarded by Studienfonds OWL
  • 2017: Admission to elite program of the EIM-faculty at Paderborn University
To top