About Me
I am Yingjie Lei, an incoming M.S.E. student in Computer Science at Johns Hopkins University in Fall 2026. Prior to this, I received dual undergraduate degrees in Artificial Intelligence from South China Normal University and the University of Aberdeen, graduating with First Class Honours from the University of Aberdeen.
My research interests lie in computer vision and world models for embodied planning and robot learning. I am especially interested in how visual and learned world representations can support action, prediction, and decision-making.
My past research experience spans test-time adaptation for medical vision, efficient video generation policy, and benchmark design for sequential decision-making agents. These projects have shaped my broader interest in building and evaluating models that connect perception, prediction, and action.
Education

Johns Hopkins University
Fall 2026M.S.E. in Computer Science
Incoming master's student in the Department of Computer Science.

South China Normal University - Aberdeen Institute of DSAI
2022-2026B.Eng. in Artificial Intelligence
Chinese degree-awarding institution in my dual-degree undergraduate program.
University of Aberdeen
2022-2026B.Sc. (Hons) in Artificial Intelligence
UK degree-awarding institution in my dual-degree undergraduate program. Graduated with First Class Honours.
Personal
Outside research, I also enjoy cycling, music, and video games. I love playing RPG games, recently including Final Fantasy VII Rebirth and Persona 5 Royal. I listen to a wide range of music, from indie rock to K-pop.
Some of the concerts I've attended can be found on my Xiaohongshu and YouTube profiles. I mostly use Xiaohongshu for non-academic posts, including travel, music, daily life, and personal interests.
News
My undergraduate thesis was selected as an Outstanding Thesis (top 10%).
Released PrefBench as an arXiv preprint.
I will join Johns Hopkins University as an M.S.E. student in Computer Science in Fall 2026.
Released Draft-and-Target Sampling for Video Generation Policy as an arXiv preprint.
Our work on test-time training for 4D medical image interpolation was accepted to IJCNN 2025 as an oral presentation.
