Zewei Zhang

I am Zewei Zhang, a PhD candidate in Electrical & Computer Engineering at McMaster University, supervised by Prof. Jun Chen. I am currently an Applied Scientist Intern at Amazon in Toronto and was a visiting PhD student at the University of British Columbia with Prof. Renjie Liao. I previously worked closely with Prof. Xiangyu Xu.

My research lies at the intersection of generative modeling, video and trajectory prediction, controllable editing, and learning-guided search. I am interested in models that reason over motion, structure, and time, and in evaluation methods that reveal how these models behave beyond short, isolated examples.

I am increasingly interested in AI systems that learn from behavioral trajectories. A trajectory, such as the evolving path of a researcher through questions, hypotheses, experiments, failures, and revisions, contains rich feedback about how actions shape outcomes over time. I view generation, evaluation, and search through this lens as an iterative feedback loop. This motivates my interest in memory, adaptation, and self-improving systems that turn experience into reusable abstractions, better control, and more reliable decision-making.

• May 2026 Joined Amazon in Toronto as an Applied Scientist Intern, focusing on human-guided video-to-video motion editing.
• Mar 2026 Released TrajLoom, a dense future trajectory generation framework for long-horizon video modeling.
• Jan 2026 Boolean Satisfiability via Imitation Learning was accepted to ICLR 2026.
• Jun 2025 Began a visiting PhD position at UBC with Prof. Renjie Liao, working on trajectory-based video generation and evaluation.
• Mar 2025 GoodDrag was accepted to ICLR 2025.