Hi, I’m Tian Xia, a master’s student in CSE at Harvard University. I recently graduated from the University of Michigan, Ann Arbor with triple majors in Honors Math, Honors CS, and DS. I previously worked as a research intern at SafoLab and SLED lab. I am deeply passionate about modeling concepts in mathematical language and examining them from different perspectives. My main excitement lies in computer vision, machine learning and robotics. Outside of research, I’m an avid gamer, enjoy long walks, and love flying to different cities.

Research

My research spans computer vision, generative modeling, and multimodal learning. I am interested in how models represent visual and spatial structure, and how we can adapt them to new tasks with useful supervision and feedback. My work includes semantic-aware 3D reconstruction, consistent video generation, and prompt optimization for multimodal clinical tasks.

More recently, I have been studying how optimization objectives and evaluation criteria shape model behavior, from ranking-aware prompt search to reflective learning in LLM agents. My current interests include:

Visual and spatial understanding: Connecting images, language, and 3D representations for perception and embodied systems.
Generative modeling: Improving video consistency and studying how synthetic data supports downstream visual learning.
Prompt optimization and evaluation: Designing feedback and selection objectives that reflect the capabilities we want models and agents to learn.

Publications

arXiv 2026

Ranking-PE overview of multimodal clinical diagnosis and ranking-aware evaluation

Ranking-PE: Ranking-Aware Prompt Optimization for Multimodal Clinical Diagnosis

Tian Xia, Minghao Liu, Yiqing Liang, Laixi Shi, Jiayun Wang

[Project][Arxiv][Paper]

3D-LLM/VLA @ CVPR 2025

SAB3R: Semantic-Augmented Backbone in 3D Reconstruction

Xuweiyi Chen*, Tian Xia*, Sihan XU, Jianing Yang, Joyce Chai, Zezhou Cheng

[Project][Arxiv][Code][Poster]

TMLR 2024

UniCtrl: Improving the Spatiotemporal Consistency of Text-to-Video Diffusion Models via Training-Free Unified Attention Control

Tian Xia*, Xuweiyi Chen*, Sihan Xu

[Project][Arxiv][Code]

Honors and Awards

2025: Outstanding Achievement in Mathematics Awards
2024: Evelyn O. Bychinsky Awards
2024: James B. Angell Scholar
2023, 2024: M.S. Keeler Department of Mathematics Merit Scholarships
2022, 2023: University Honors

Educations

2025.08 - now, S.M. in Computational Science and Engineering, Harvard University
2022.09 - 2025.05, B.S. in Honors Mathematics, Honors Computer Science and Data Science, University of Michigan, Ann Arbor