Kaisen Yang

Undergraduate Researcher · Tsinghua University  ·  yks23@mails.tsinghua.edu.cn

kaisen.jpg

(+86) 150 1900 0811

Beijing, China

Biography

I am a third-year undergraduate at the Department of Computer Science and Technology, Tsinghua University, with a GPA of $\text{3.9/4.0}$, ranking $\text{9th}$ out of $\text{171}$ students.

I am broadly fascinated by Generative Models — how we train them, structure them, and stress-test them. Concretely, my current work spans three layers:

Layer 1 · Paradigms

Modeling Paradigms

Masked and discrete diffusion, structured reasoning, and the boundary between autoregressive and non-autoregressive generation.

Layer 2 · Systems

Architectures & Efficiency

Diffusion-Transformer quantization, sampler design, and attention mechanisms for faster, more reliable generation.

Layer 3 · Evidence

Data & Benchmark

Datasets and benchmarks that expose what generative models can do, where they fail, and why the failure matters.

I care less about chasing a single architecture, and more about the question: what is the smallest set of inductive biases (paradigm + architecture) that, given the right data, gets us a robust generative model?

Research Affiliations

Now ·
Tsinghua C3I Lab — Center for Collaborative & Conversational Intelligence, Tsinghua University. I am advised by Ning Ding and Prof. Bowen Zhou.
2026.05 –
Nature Will — Algorithm Intern.
2025.12 –
Einsia.AI — Lab Leader.
Past ·
Lamb Group — student researcher with Prof. Alex Lamb, on masked diffusion and structured reasoning.

Service & Activities

  • Chairman, Students’ Association of Science and Technology, Department of CST, Tsinghua University.
  • Council Member, 19th Spark Program (星火计划), Tsinghua University.
  • Former Class President, Class 31, Department of CST, Tsinghua University.

news

Jun 17, 2026 🏆 Awarded the SenseTime Scholarship 2026.
Jun 16, 2026 ✍️ Our paper, VoidPadding, is released on arXiv.
Jun 02, 2026 I became Chairman of SAST of the Computer Science and Technology (CST) department.
May 01, 2026 🎉 2 papers have been accepted to ICML 2026!
Feb 15, 2026 🎉 Our paper, PAM, has been accepted to CVPR 2026!
Feb 01, 2026 ✍️ 1 preprint released on arXiv.

selected publications

  1. ArXiv
    browserbc.png
    Scalable Behaviour Cloning on Browser Using via Skill Distillation
    Kaisen Yang, Zheng Jiang, Yuzhao Peng, Houde Qian, Boshi Zhang, and 9 more authors
    2026
    arXiv HTML PDF
  2. ArXiv
    voidpadding.png
    VoidPadding: Let [VOID] Handle Padding in Masked Diffusion Language Models so that [EOS] Can Focus on Semantic Termination
    Chunyu Liu, Zhengyang Fan, Kaisen Yang, and Alex Lamb
    arXiv preprint arXiv:2606.17999, 2026
    arXiv PDF Code
  3. ArXiv
    frontier-eng.png
    Frontier-Eng: Benchmarking Self-Evolving Agents on Real-World Engineering Tasks with Generative Optimization
    Yizhe Chi, Deyao Hong, Dapeng Jiang, Tianwei Luo, Kaisen Yang, and 6 more authors
    arXiv preprint arXiv:2604.12290, 2026
    arXiv HTML PDF Code
  4. ICML
    masked-diffusion.png
    Improving Sampling for Masked Diffusion Models via Information Gain
    Kaisen Yang, Jayden Teoh, Kaicheng Yang, Yitong Zhang, and Alex Lamb
    In Proceedings of the 43rd International Conference on Machine Learning, 2026
    arXiv HTML PDF Code
  5. CVPR
    pam.png
    PAM: A Pose–Appearance–Motion Engine for Sim-to-Real HOI Video Generation
    Mingju Gao†, Kaisen Yang†, Huan-ang Gao, Bohan Li, Ao Ding, and 11 more authors
    In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
    arXiv HTML PDF
  6. ICLR
    cubebench.png
    CubeBench: Diagnosing Interactive, Long-Horizon Spatial Reasoning Under Partial Observations
    Huan-ang Gao, Zikang Zhang, Tianwei Luo, Kaisen Yang, Xinzhe Juan, and 7 more authors
    In International Conference on Learning Representations (ICLR), 2026
    arXiv HTML PDF
  7. ACL
    tree-kg.png
    Tree-KG: An Expandable Knowledge Graph Construction Framework for Knowledge-intensive Domains
    Songjie Niu†, Kaisen Yang†, Rui Zhao, Yichao Liu, Zonglin Li, and 2 more authors
    In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2025
    HTML PDF
  8. NeurIPS
    impromptu-vla.png
    Impromptu VLA: Open Weights and Open Data for Driving Vision-Language-Action Models
    Haohan Chi, Huan-ang Gao, Ziming Liu, Jianing Liu, Chenyu Liu, and 6 more authors
    Advances in Neural Information Processing Systems (NeurIPS), 2025
    arXiv HTML PDF Code