Xun Wang
About me
I am a third-year Ph.D. student in Computer Science at the Technical University of Munich (TUM) and CISPA Helmholtz Center for Information Security, supervised by Prof. Michael Backes. I work daily with Franziska Boenisch and Adam Dziedzic at SprintML Lab.
My research focuses on trustworthy and efficient machine learning, including privacy-preserving model adaptation and post-training, model provenance and ownership, and trustworthy world models. I study what foundation models reveal about private data and how to trace the origins of models and agents.
I received my M.Sc. in Robotics, Cognition, Intelligence from TUM in 2023, graduating with high distinction, and my B.Eng. in Vehicle Engineering from Tongji University in 2020.
Research interests
- Privacy auditing and protection. Measuring privacy leakage in foundation-model embeddings and LLM adaptations, and developing privacy-preserving prompt transfer methods.
- Model provenance and ownership. Auditing the identity of agentic LLM APIs, fingerprinting generative models, and protecting model intellectual property.
- Adaptation and post-training. Privacy-preserving and efficient adaptation of foundation models, alongside architecture-agnostic structured pruning.
- Trustworthy world models. Tracing whether an agent was trained in a particular world model.
See my publications for papers and project pages, or my CV for education and experience. You can also download my CV (PDF).
News
-
Our paper “Auditing Privacy Leakage in Tabular Foundation Model Embeddings” was accepted to NeurIPS 2026.
-
AgentProv, our work on auditing agentic LLM API providers, was accepted to EMNLP 2026.
-
“Benchmarking Empirical Privacy Protection for Adaptations of Large Language Models” was accepted to ICLR 2026 as an oral presentation.