张泽楷 (Zekai Zhang)

About Me

I am a third-year PhD student in EECS at the University of Michigan, advised by Prof. Qing Qu. Before that, I received my bachelor’s degree from the Department of Automation, Tsinghua University.

Research

I work on effective theory for generative models (especially diffusion models), via representation learning and simplified models. My goal is to help people understand, disenchant, and collaborate with generative models better.

News

• Oct 2026
  One paper on diffusion language models is on arXiv!

  Scaling and Distilling Text Embeddings for Better Diffusibility
• Aug 2026 Gratefully wrapped up my internship at TikTok Seattle (May–Aug).
• May 2026
  One paper accepted at ICML 2026, and a Silver Reviewer award.

  Evaluating the Representation Space of Diffusion Models via Self-Supervised Principles
• Jan 2026
  One paper accepted at ICLR 2026.

  Generalization of Diffusion Models Arises with a Balanced Representation Space, with an oral talk at DeepMath.
• Oct 2025
  Two papers accepted at NeurIPS 2025.

  A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspective

  Understanding Representation Dynamics of Diffusion Models via Low-Dimensional Modeling
• May 2025 Top Reviewer award at ICML 2025.