01
World Models and Video Generation
Physically plausible generation, multi-shot visual reasoning, controllable motion, and video-native representations.
HKUST AMC
DNA: Digital Neural Alive
DNA Lab studies generative world models, multimodal agents, and controllable visual intelligence for systems that can see, make, reason, and act in dynamic environments.
What We Do
We build AI systems around three connected questions: how neural models represent a changing world, how humans steer generated media and agents, and how research prototypes become reliable tools for creative, scientific, and social use.
Research Themes
DNA Lab's agenda connects image and video generation, world modeling, embodied and GUI agents, efficient model adaptation, and human-centered applications.
02
Multimodal Agents
Agents that ground decisions in pixels, structure, evidence, and interactive feedback instead of brittle language-only traces.
03
Efficient Neural Systems
Quantization, feature reuse, preference optimization, and scalable training or inference methods for modern generative models.
04
Human and Creative AI
AI-guided creation, narrative self-films, health companions, video tools, and interfaces that keep human intent in the loop.
Who We Are
DNA Lab is led by Harry Yang at HKUST AMC, with students and interns working across visual generation, multimodal reasoning, agents, and applied AI systems.
Harry Yang
Assistant Professor at HKUST AMC and affiliate of HKUST Von Neumann Institute. Previously Visiting Professor at Yale and Research Scientist at Meta.
Current Students
Zihao WangHKUST
Yexin LiuHKUST-GZ
Xuran MaBIT
Mingzhe ZhengNPU
Jiajun ZhaFudan
Yaofu LiuHKUST
Haoze ZhengBeijing Jiaotong
Wenjie ShuUESTC
Weixiao WangCUHK
Dengyang JiangNPU
Haiying HeChina Agricultural University
Selected Publications
Recent work from Harry Yang's group, selected from the current hyang.org research list.
RotateAttention: RoPE-Aware Rotation and Range Rectification for INT4
Quantized Attention in Video Generation
Do GUI Agents Believe Their Eyes? Diagnosing State-Belief Reliance on
Pixels versus Structure
AlignVid: Training-Free Attention Scaling for Semantic Fidelity in
Text-Guided Image-to-Video Generation
VideoGen-of-Thought: Step-by-Step Generation of Multi-Shot Videos
Make-A-Video: Text-to-Video Generation without Text-Video Data
Contact and Join
DNA Lab welcomes strong PhD, RA, and internship candidates interested in video generation, multimodal agents, world models, and AI systems.
Affiliation
Openings
RPG students, research assistants, and interns. Strong students are encouraged to apply for HKPFS.