PhD
Chair of Data Analytics and Machine Learning at TUM
Informatik 26
Department of Computer Science
Boltzmannstr. 3
85748 Garching
Biosketch
Yan is a PhD student in the Data Analytics and Machine Learning group at the Technical University of Munich (TUM) under the supervision of Prof. Stephan Günnemann. His research interests include robust machine learning and machine learning on graphs.
Yan started his PhD in 2022. Before that, he graduated from TUM in Informatics (M.Sc.) with high distinction, and from Paderborn University in Computer Science (B.Sc.) with a Minor in Mathematics (with distinction).
You can find out more about him on his website.
relAI Research
Trustworthy Machine Learning for Large Language Models
My research focuses on developing trustworthy and reliable machine learning methods for large language models (LLMs), with particular emphasis on unlearning, alignment, and adversarial robustness. Recent work introduces a probabilistic evaluation framework for LLM unlearning and alignment, demonstrating that deterministic evaluations can substantially overestimate safety and unlearning performance. Building on this, we develop new methods for machine unlearning that leverage model collapse to remove information while preserving utility. I am also interested in studying adversarial robustness in generative models, in a recent work we showed that standard single-generation evaluations underestimate vulnerabilities of LLMs under adversarial attacks. Overall, my research aims to advance principled foundations and practical tools for building safe, reliable, and trustworthy AI systems.
Publications
https://arxiv.org/pdf/2507.04219
https://arxiv.org/pdf/2507.04446
https://arxiv.org/pdf/2410.03523
https://arxiv.org/pdf/2410.09878