Xiaoyu (Nicholas) Wu · 吴晓宇
Hi, I'm Xiaoyu Wu
PhD student · Carnegie Mellon University
I work on trustworthy generative AI: privacy and copyright risks in diffusion and language models, and methods that make generative systems more reliable.
News
- new Released DSR, our work on taming outlier tokens in diffusion transformers.
- new Our work on the corruption stage in few-shot diffusion fine-tuning was accepted to KDD 2026.
- Our work on extracting forgotten data after exact unlearning was accepted to NeurIPS 2025.
- FineXtract, our training-data extraction work for personalized diffusion models, was accepted to ICML 2025.
- Our team won all four tracks of the MIDST Challenge at SaTML 2025.
Selected work
Preprint · 2026
Taming Outlier Tokens in Diffusion Transformers
Dual-Stage Registers reduce harmful outlier tokens in both vision encoders and diffusion transformers, improving generation quality across ImageNet and text-to-image settings.
ICML · 2025
Leveraging Model Guidance to Extract Training Data from Personalized Diffusion Models
FineXtract uses the distribution shift between pretrained and personalized diffusion models to recover about 20% of fine-tuning data from real-world checkpoints.
About
I'm a PhD student at Carnegie Mellon University, where I work with Prof. Niloofar Mireshghallah, Prof. Zhiwei Steven Wu, and Prof. Andrew Ilyas. My research spans data extraction, machine unlearning, membership inference, copyright protection, and the robustness of generative models. I also collaborate with Prof. Chen Wei on generative modeling.
I co-founded a volunteer group focused on copyright issues surrounding image generation. We build open-source tools and provide technical support for AI copyright litigation.