Portrait of Yuheng Wu

Yuheng Wu

Ph.D. Student

I’m a PhD student in the School of Computing at the Korea Advanced Institute of Science & Technology (KAIST). I work with my colleagues at the Collaborative Distributed Systems and Networks Lab under the supervision of Prof. Dongman Lee. My research interests lie in building efficient and real-time interactive video world models.

Education

KAIST logo

Korea Advanced Institute of Science and Technology

Sep 2025 — 2028 (expected)

Ph.D. in School of Computing

KAIST logo

Korea Advanced Institute of Science and Technology

Sep 2023 — Jul 2025

M.S. in School of Computing

Shanghai University logo

Shanghai University

2019 — 2023

B.Eng. in Computer Science & Cyber Security

Internship

Visko AI logo

Visko AI

Dec 2025 — Present

Working on autoregressive distillation for the Orbis model.

News

My first author paper “Background Fades, Foreground Leads: Curriculum-Guided Background Pruning for Efficient Foreground-Centric Collaborative Perception “ has been accepted by ICRA 2026! We release all the code, checkpoints (including all baselines) and visulizations for reproducibility! Welcome to fork!
My first author paper “How2Compress: Scalable and Efficient Edge Video Analytics via Adaptive Granular Video Compression” has been accepted by ACM MM2025!
Langcoop is accepted by CVPR 2nd Workshop on Multi-Agent Embodied Intelligent Systems 2025. Thanks Xiangbo and all collaborators! Check our paper and code.

Selected Publications

  1. arXiv
    Delta Forcing: Trust Region Steering for Interactive Autoregressive Video Generation
    Yuheng Wu, Xiangbo Gao, Tianhao Chen, Xinghao Chen, Qing Yin, Zhengzhong Tu, and Dongman Lee
    arXiv, 2026
  2. ACM MM
    how2compress.png
    How2Compress: Scalable and Efficient Edge Video Analytics via Adaptive Granular Video Compression
    Yuheng Wu, Thanh-Tung Nguyen, Lucas Liebe, Nhat-Quang Tau, Pablo Espinosa Campos, Jinghan Cheng, and Dongman Lee
    ACM Multimedia, 2025
  3. ICRA
    fadelead.png
    Background Fades, Foreground Leads: Curriculum-Guided Background Pruning for Efficient Foreground-Centric Collaborative Perception
    Yuheng Wu, Xiangbo Gao, Quang Tau, Zhengzhong Tu, and Dongman Lee
    IEEE International Conference on Robotics and Automation, 2026