Zekun Cai

I am a postdoctoral researcher at the University of Tokyo. I obtained my Ph.D. in 2024 from the Center for Spatial Information Science at the University of Tokyo, where I was supervised by Prof. Ryosuke Shibasaki and Prof. Xuan Song. Additionally, I conducted a research visit at Emory University from 2023. Prior to that, I received my M.S. from the University of Tokyo in 2021 and my B.S. from the University of Electronic Science and Technology of China in 2018.

During my Ph.D., I was fortunate to be co-advised by Prof. Renhe Jiang at UTokyo and collaborated with Prof. Liang Zhao at Emory University. My work centered on developing principled approaches for temporal domain generalization under continuously evolving data distributions. In parallel with my academic research, I also collaborated with industry partners such as Tencent and Yahoo! JAPAN, focusing on analyzing, modeling, and predicting trillion-scale human behavioral data in real-world environments.

I have published over 20 peer-reviewed papers in venues including NeurIPS, KDD, TKDE, TMC, and CIKM, and I regularly serve as a PC member for conferences such as NeurIPS, KDD, IJCAI, and CIKM.

Research Interests

My recent research focuses on developing mathematically and physically grounded approaches to enhance the adaptability and generalization of AI systems in complex environments. Key areas include:

• Continuous Domain Generalization: Synchronizing evolution of data and neural networks through Structured Representations, Manifold Geometry, Group Equivariant, Neural Differential Equations, etc.
• Spatiotemporal Learning at Scale: Graph Neural Network, Spatiotemporal Foundation Models, Dynamic Representation Learning, Real-world Applications, etc.