Yuzheng Xu

Research Interests

I work on how we tell whether a language model is actually good at something, and on what happens when a person and a model have to reach an answer together rather than one replacing the other — evaluation reliability, and human–AI complementarity. Before that, computer vision: video understanding and annotation, and interfaces that give people real-time feedback on their own performance.

Selected Publications

• Toward Human-AI Complementarity Across Diverse Tasks Yuzheng Xu, A. Dahmani, M. D. Blanchard, et al. Under review · arXiv:2605.04070
• Am I More Pointwise or Pairwise? Revealing Position Bias in Rubric-Based LLM-as-a-Judge Yuzheng Xu, T. Hirasawa, T. Kozuno, Y. Ushiku Under review · arXiv:2602.02219
• MK2 at PBIG Competition: A Prompt Generation Solution Yuzheng Xu, T. Hirasawa, S. Kawano, S. Kato, T. Kozuno IJCAI AgentScen Workshop 2025 · ACL Anthology · arXiv
• VisDev: A Stroke Deviation Visualization System Using Point-Set Registration for Handwriting Training Yuzheng Xu, S. Yoshida, A. Hashimoto, Y. Ushiku CHI Late-Breaking Work 2025 · ACM DL
• Video Region Annotation with Sparse Bounding Boxes Yuzheng Xu, Y. Wu, N. S. Zuraimi, S. Nobuhara, K. Nishino IJCV 2023 · BMVC 2020 (Best Student Paper) · Springer · arXiv

Keywords

LLM evaluation · Human–AI complementarity · Self-awareness · AI for science