Mingjing '1mage' Yi
M.S. in Computer Science · Columbia University
Research Assistant, Mobile X Lab · Research Intern, SeeleAI · Applied Scientist Intern, Microsoft
I work on multimodal models and generative AI, with a recent focus on 3D representation and 3D reconstruction.
About
I'm a Computer Science M.S. student at Columbia University, currently a Research Assistant at the Mobile X Lab, co-advised by Prof. Xia Zhou and Prof. Changxi Zheng. My research centers on multimodal models and generative AI, with a recent focus on 3D representation and 3D reconstruction. I'm also an Applied Scientist Intern at Microsoft (Copilot Tuning) and a Research Intern at SeeleAI. I enjoy building models and systems that connect perception across vision, language, and 3D.
Research Interests
Publications
Experience
May 2026 – Now
Applied Scientist Intern · Microsoft
Redmond, WA. On the Frontier Tuning team: built an automated bug-filing skill for the agentic AI harness shipped in GitHub Copilot; benchmarked Copilot Tuning levers across agentic AI workflows on enterprise datasets, evaluating test-time compute and tuning methods from context and orchestration to RL.
Sept 2025 – Now
Research Assistant · Mobile X Lab, Columbia University
New York, NY. Co-advised by Prof. Xia Zhou and Prof. Changxi Zheng. Developed RollingPol, a single-shot computational imaging system with a VAE-based reconstruction network extracting physics-based polarization priors for dynamic underwater image enhancement (MobiCom 2026). Currently working on 3D reconstruction for low-light, low-visibility underwater scenes.
May 2025 – Now
Algorithm Research Intern · SeeleAI
Shenzhen, CN. Co-first author on EVA01, a native 3D large multimodal model with a Mixture-of-Transformers (MoT) architecture unifying 3D understanding, text-to-3D generation, and context-aware editing (SIGGRAPH Asia 2026, under review). Also built SOTA text-to-motion generation and expanded a ~20k-clip motion dataset.
May 2024 – Aug 2025
Research Assistant · SMIIP Lab, DKU
Kunshan, CN. Advised by Prof. Ming Li. Built a diffusion-based video-to-audio model with visual scene detection (CLIP/CLAP); scene detection accuracy improved by 50%+ and generation quality by 24% over SOTA (APSIPA ASC 2025). Explored Diffusion Transformer (DiT) with ControlNet for temporal alignment.
2023 – 2024
Teaching Assistant · Duke Kunshan University
Kunshan, CN. TA for Intro to Data Science and Intro to Programming and Data Structures; held 50+ office hours and supported recitation sessions.
Education
2025 – 2026
M.S. in Computer Science · Columbia University
New York, NY · GPA 4.0/4.0. Coursework: LLM-based GenAI, Deep Learning, Computer Graphics, NLP, Continual Learning and Memory Models.
2021 – 2025
B.S. in Computer Science · Duke Kunshan University & Duke University
Kunshan, CN & Durham, NC · GPA 3.7/4.0. Dual B.S.: Computer Science (DKU) and Interdisciplinary Studies in Computer Science (Duke). Honors: Entrance Scholarship (75%), Dean's List (4x).
Skills
Languages: Chinese (native), English.
Contact
Feel free to reach out via email. I'm happy to chat about research and collaboration.