Frozen in a Frame: The Velocity Blind Spot in JEPA World Models
Zhang et al., 2026.
CV / Curriculum Vitae
A printable PDF version is available here: Download CV.
Education
- IIIS (Yao Class), Tsinghua University, B.S. in Computer Science and Technology, Beijing, China. 2024-2028 (expected).
- GPA: 3.9 / 4.0 · Rank: 9 / 92.
- Selected coursework: Deep Learning, Machine Learning, Natural Language Processing, Embodied Artificial Intelligence, Advanced Computer Graphics, Introduction to Large Language Model Applications, and Type-safe Modern System Practice.
Research interests
- Efficient attention and architecture design for language models and diffusion transformers
- Masked diffusion language models and controllable generation
- Activation quantization and hardware-aware training
- Long-context memory and retrieval
- Multimodal and embodied agents
Research experience
- Undergraduate Research Intern, Tsinghua College AI · Mar 2026 – Present. Supervisor: Alex Lamb. Research on VoidPadding, JustQuant, and CoMem.
- Undergraduate Research Intern, Tsinghua SAIL Group · Jun 2025 – Feb 2026. Supervisors: Jianfei Chen and Jun Zhu. Efficient attention for LLMs and DiTs.
Industry experience
- Research Intern, Ubiquant IQuest Lab · Jun 2026 – Sep 2026. LLM pretraining algorithms and DeepSeek Sparse Attention.
- Technical Staff, Einsia · Aug 2026 – Present. AgentGit search and VideoCut Bench.
Publications
-
JustQuant: You Don’t Need Smoothing, SVD, or Rotation for 4-Bit Activation Quantization
Yang et al., 2026.
-
Understanding Is Done Early: A Depth Division of Labor in Large Language Models and Its Use for Unbounded-Context Memory
Liu et al., 2026.
Selected projects
- Moe Traits 2 Companion — multimodal AI desktop companion with tool-calling agents and persistent memory.
- Efficient Attention for Diffusion Transformers — NlogNAttn for LightningDiT.
- PIANO — character-aware AI vocal-cover pipeline.
- HACT — hierarchical ACT policy for long-horizon embodied tasks.
Awards
- Silver Medal, 21st Chinese Girls’ Mathematical Olympiad · 2022
- Freshman Scholarship, Tsinghua University · 2024–2025
- Artistic Excellence Scholarship, Tsinghua University · 2024–2025
Skills
- Programming: Python, C/C++, Scala, TypeScript, HLSL, Bash, LaTeX, Rust
- Tools: PyTorch, Hugging Face Transformers, Linux, Git, Slurm, Docker
- Languages: Chinese (native), English (TOEFL 108)