Xuefeng Hu
Portrait of Xuefeng Hu

Xuefeng Hu

Member of Technical Staff·AMI Labs·Paris

I am building world models with real-world understanding at AMI Labs (Advanced Machine Intelligence) in Paris.

Before that, I spent two and a half years at ByteDance / TikTok, most recently as a Staff Research Scientist leading video foundation model pretraining. Earlier there I worked on unified understanding-and-generation models, multimodal reinforcement learning, scalable memory and multimodal LLMs for math. I received my Ph.D. in Computer Science from the University of Southern California, advised by Prof. Ram Nevatia, where I studied how to adapt pre-trained vision–language models to new domains with little or no labeled data.

  • World Model
  • Memory, Continual Learning and Test-time Adaptation
  • Unified Understanding and Generation Models
  • Multimodal RL

News

  • Sep 2026

    BitDance accepted to NeurIPS 2026.

  • Aug 2026

    Intervention-Certified Amplification accepted to EMNLP 2026 Findings.

  • Jul 2026

    Joined AMI Labs in Paris as a Member of Technical Staff, building world models.

  • Jun 2026

    Released the ARM technical report, an autoregressive multimodal model with unified discrete representations.

  • Mar 2026

    Promoted to Staff Research Scientist at ByteDance / TikTok.

  • Feb 2026

    Implicit Neural Representation Facilitates Unified Universal Vision Encoding accepted to CVPR 2026.

  • Sep 2025

    DiCo accepted to NeurIPS 2025 as a Spotlight.

  • Aug 2025

    InfiMM-WebMath-40B accepted to EMNLP 2025 Findings; the dataset has 16K+ downloads on Hugging Face.

  • Feb 2024

    Joined ByteDance Seed-Foundation as a research scientist.

  • Jan 2024

    ReCLIP presented as an oral at WACV 2024 and selected as a Best Paper Finalist.

  • Dec 2023

    Completed my Ph.D. at USC.

Publications

Preprints

Selected papers

* Equal contribution. Full list on Google Scholar.

Experience

  • 2026 – now
    AMI Labs · Member of Technical Staff Advanced Machine Intelligence · Paris, France

    World model research.

  • 2024 – 2026
    ByteDance / TikTok San Jose, CA · Bellevue, WA
    • Staff Research ScientistMar 2026 – Jul 2026

      Led video foundation model pretraining.

    • Senior Research ScientistJul 2025 – Mar 2026

      Unified understanding and generation models; multimodal RL; scalable memory.

    • Research ScientistNov 2024 – Jun 2025

      Agentic RL for multimodal LLMs.

    • Research Scientist, Seed-FoundationFeb 2024 – Oct 2024

      Multimodal LLMs for math: data, continued pretraining, SFT and RL.

  • 2022
    Amazon Lab126 · Research Intern Bellevue, WA

    Developed ReCLIP, improving CLIP’s average zero-shot accuracy by 5.11% across 22 benchmarks without extra supervision.

  • 2021
    Meta · Research Intern Boston, MA

    Online test-time adaptation of pre-trained models under distribution shift.

  • 2020
    Facebook · Research Intern Boston, MA

    Part-level contrastive learning for weakly-supervised fine-grained anomaly detection.

Education

  • 2018 – 2023
    University of Southern California · Ph.D., Computer Science Advisor: Ram Nevatia · Dissertation: Adapt Pre-trained Representation Towards Downstream Tasks
  • 2017 – 2018
    Columbia University · M.S., Computer Science Advisor: Shih-Fu Chang
  • 2015 – 2017
    University of Michigan, Ann Arbor · B.S.E., Computer Science Advisors: Jia Deng, Yaoyun Shi
  • 2013 – 2017
    Shanghai Jiao Tong University · B.S.E., Electrical & Computer Engineering

Selected honors

  • 2024

    Best Paper Award Finalist, WACV 2024

  • 2022

    “30 under 30”, All America Chinese Youth Federation

  • 2017

    James B. Angell Scholar, University of Michigan

  • 2012

    National First Prize, China High School Mathematics Competition

Photography

Street photography from the cities I have lived in and passed through.

  • Vintage car in black and white, 2019
  • Beach at sunset, 2020
  • Cyclists at golden hour, 2019
  • Orange umbrella in the rain, 2026
  • Storefront at night, 2026
  • Woman walking down a narrow alley, 2023
  • Taxis at night, 2023
  • Crosswalk with a passing bus, 2023
  • Neon-lit market, 2023

More on Instagram, @xuefenghu.

Beyond research

In 2019 I founded LA Stories (洛杉矶文化沙龙), a non-profit public forum in Los Angeles that hosts talks, panel discussions and film screenings for the local Mandarin-speaking community, with more than 2,000 registered audience members.

Away from research, I am happiest around a table: cooking, hosting dinners that bring people together, and mixing cocktails at Midnight Choir, my make-believe home bar. I play guitar and djembe and love a good jam session. I keep returning to film, photography, painting, writing, mathematics, philosophy and Burning Man, and I am a cat person through and through.

Along the way I have been shaped by the words of Constantine P. Cavafy, Friedrich Nietzsche and 苗炜; the music of Leonard Cohen, Pink Floyd, 张悬 / 安溥, 野孩子, 张玮玮 & 郭龙 and 万能青年旅店; and the talk show 锵锵三人行.

Home has been Dalian, Shanghai, Ann Arbor, New York, Los Angeles, Seattle, San Jose, and now Paris.

As you set out for Ithaka, hope the voyage is a long one.

C. P. Cavafy, Ithaka