photo_me.jpg

Kaiwen Zheng (郑凯文)

Ph.D. student, TSAIL Group

Department of Computer Science & Technology, Tsinghua University (China)


I’m a fourth-year Ph.D. student at the Department of Computer Science and Technology, Tsinghua University, advised by Prof. Jun Zhu. Before that, I received my B.E. degree from the same department at Tsinghua University in 2022. In the summer of 2024, I was honored to have an internship at NVIDIA Deep Imagination Research in the San Francisco Bay Area.

My research focuses on developing principled, insightful, scalable, efficient and effective training/inference techniques for deep generative models, with a particular emphasis on diffusion-related models. I am also interested in reinforcement learning, unified multimodal model and world model.

Selected Publications & Preprints [full list]

(*) denotes equal contribution; (†) denotes corresponding author

  1. TurboDiffusion: Accelerating Video Diffusion Models by 100-200 Times
    Jintao Zhang*Kaiwen Zheng*, Kai Jiang* , Haoxu Wang*, Ion Stoica, Joseph E Gonzalez, Jianfei Chen, and Jun Zhu
    Technical Report, 2025
  2. World Simulation with Video Foundation Models for Physical AI
    NVIDIA
    Technical Report, 2025
  3. Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency
    In The Fourteenth International Conference on Learning Representations, 2026
  4. DiffusionNFT: Online Diffusion Reinforcement with Forward Process
    In The Fourteenth International Conference on Learning Representations, 2026

    Oral, Top 1.2%

  5. Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator
    In Proceedings of the 42nd International Conference on Machine Learning, 2025

    Spotlight, Top 2.6%

  6. Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling
    In The Thirteenth International Conference on Learning Representations, 2025

    Top 5%

  7. Diffusion Bridge Implicit Models
    Kaiwen Zheng*Guande He*Jianfei ChenFan Bao, and Jun Zhu
    In The Thirteenth International Conference on Learning Representations, 2025
  8. Vidu: a Highly Consistent, Dynamic and Skilled Text-to-Video Generator with Diffusion Models
    Fan BaoChendong Xiang*, Gang Yue*Guande He*Hongzhou Zhu*Kaiwen Zheng*Min Zhao*Shilong Liu* , Yaole Wang*, and Jun Zhu
    Technical Report, 2024
  9. DPM-Solver-v3: Improved Diffusion ODE Solver with Empirical Model Statistics
    Kaiwen Zheng*Cheng Lu*Jianfei Chen, and Jun Zhu
    In Advances in Neural Information Processing Systems, 2023
  10. Improved Techniques for Maximum Likelihood Estimation for Diffusion ODEs
    Kaiwen Zheng*Cheng Lu*Jianfei Chen, and Jun Zhu
    In Proceedings of the 40th International Conference on Machine Learning, 2023

Experience


Deep Imagination Research, NVIDIA

Research Intern

2024.06 - 2024.11

Santa Clara, California, USA

Shengshu Technology

Research Intern

2023.03 - 2024.05

Beijing, China

Y-tech, Kuaishou Technology

Engineering Intern

2021.06 - 2021.08

Beijing, China

Miscellaneous

My slides for TSAIL reading group:

Personal

  • I enjoy playing ping-pong🏓 in my free time. I also watch esports (especially League of Legends) and have a taste for Chinese calligraphy (a piece of the poem 'Qingming' ("清明" in Chinese)).
  • I customized and hosted a Minecraft server at middle school ([Pic1][Pic2][Pic3][Pic4]).
  • I am enthusiastic about math since middle school. In high school, I compiled and wrote a set of interesting math exercises to form a test ([Problems][Answers], in Chinese).