JJJYmmm

Jie Huang

Institute of Computing Technology, Chinese Academy of Sciences
Beijing, China · 1650675829 [at] qq [dot] com

I am an M.Eng. candidate at the Institute of Computing Technology, Chinese Academy of Sciences. My research focuses on multimodal architectures, efficient inference, and agents, from vision-language models to multimodal agent systems and on-device robotics.

I am a research intern with the Qwen Team at Alibaba Cloud, a core contributor to Qwen3-VL and Qwen3.5, and the maintainer of Qwen-MM-Plugins. I enjoy building open-source tools that bring multimodal models into practical use.

Google Scholar GitHub Website
Portrait of Jie Huang

News

Experience

Institute of Computing Technology, CAS
M.Eng. candidate · Sep. 2024 - Jun. 2027 (expected)
GPA: 3.89/4.0
Huazhong University of Science and Technology
B.Eng. · Sep. 2020 - Jun. 2024
GPA: 3.94/4.0
Qwen Team, Alibaba Cloud
Research intern · Mar. 2025 - Present
Core contributor to Qwen3-VL and Qwen3.5, working on multimodal model architecture, training/inference implementation, and open-source release.
Architecture work includes MRoPE-Interleave, DeepStack, timestamp-aligned video modeling, and Vision MTP.
Supported day-zero model integration across Transformers, vLLM, SGLang, llama.cpp, and MLX.

Selected Publications & Technical Writing

  1. Qwen3-VL Technical Report
    Core contributor
    arXiv preprint, 2025.
    Contributed to model architecture, training/inference implementation, and the open-source release.
    PaperCode
  2. Qwen3.5: Towards Native Multimodal Agents
    Core contributor
    Technical blog, 2026.
    Contributed to vision-language model architecture and inference.
    BlogCode
  3. Qwen3.8-Omni: Towards Native Omni-Modal Agents
    Contributor · Maintainer of Qwen-MM-Plugins
    Technical report, 2026.
    PaperQwen-MM-Plugins
  4. Revisiting Multimodal Positional Encoding in Vision-Language Models
    Jie Huang*, Xuejing Liu*, Sibo Song, Ruibing Hou, Hong Chang, Junyang Lin, Shuai Bai
    International Conference on Learning Representations (ICLR), 2026.
    Proposed MHRoPE and MRoPE-Interleave for spatial and long-video understanding. MRoPE-Interleave was adopted in Qwen3-VL and Qwen3.5.
    PaperCode
  5. RefHCM: A Unified Model for Referring Perceptions in Human-Centric Scenarios
    Jie Huang, Ruibing Hou, Jiahe Zhao, Hong Chang, Shiguang Shan
    IEEE Transactions on Multimedia (TMM), 2026.
    Developed a unified autoregressive model for human-centric referring perception.
    PaperCode
  6. Stealthy and Effective Physical Adversarial Attacks in Autonomous Driving
    Man Zhou, Wenyu Zhou, Jie Huang, Junhui Yang, Minxin Du, Qi Li
    IEEE Transactions on Information Forensics and Security (TIFS), 2024.
    Paper

Open Source