Soonwoo Kwon

I’m a research engineer at NAVER developing shopping foundation models. I did my Master’s degree at KAIST, under the supervision of Prof. Eunho Yang.

My research interest lies at the intersection of world models and embodied agents. Specifically:

  • Closing the sim-to-real gap through world models that incorporate 3D structure and long-horizon scene consistency.
  • Developing generalized and scalable foundation models for embodied intelligence, capable of adapting across tasks, modalities, and environments.
  • Enabling System 2 reasoning capabilities in embodied agents by utilizing world models for lookahead planning.

My prior work spans generative models([C5], [C6], [P1]), multi-modal learning([C4], [J1], [T1], [T2]), and interactive systems([C1], [C2], [C3]), forming the methodological basis for my current research direction.

Work Experiences

  • Research Engineer, NAVER, Gyeonggi-do, South Korea (Apr 2025 - Present)
  • Research Scientist, Twelve Labs, Seoul, South Korea (Apr 2024 - Apr 2025)
  • Research Scientist, Riiid, Seoul, South Korea (Mar 2022 - Apr 2024)
  • Research Intern, Socar, Seoul, South Korea (Oct 2021 - Feb 2022)
  • Research Intern, KIXLAB, Daejon, South Korea (Jan 2019 - May 2019)
  • Research Intern, Dutt’s Research Group(DRG), Irvine, USA (Jun 2018 – Dec 2018)

Preprints

     
  • [P1] Toward Stable World Models: Measureing and Addressing World Instability in Generative Environments [paper]
        Soonwoo Kwon*, Jin-Young Kim*, Hyojun Go, Kyungjune Baek, Arxiv 2025  

Journal Publications

     
  • [J1] ScoreCL: Augmentation-Adaptive Contrastive Learning via Score-Matching Function [paper]
      Jin-Young Kim*, Soonwoo Kwon*, Hyojun Go*, Yunsung Lee, Seungtaek Choi, Machine Learning 2024  

Conference Publications

     
  • [C6] SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis [paper]
      Hyojun Go*, Byeongjun Park*, Jiho Jang, Jin-Young Kim, Soonwoo Kwon, Changick Kim, CVPR 2025  
    •  
  • [C5] Denoising Task Difficulty-based Curriculum for Training Diffusion Models [paper]
      Jin-Young Kim*, Hyojun Go*, Soonwoo Kwon*, Hyun-Gyoon Kim, ICLR 2025  
    •  
  • [C4] BIPED: Pedagogically Informed Tutoring System for ESL Education [paper]
      Soonwoo Kwon, Sojung Kim, Minju Park, Seunghyun Lee, Kyuseok Kim, ACL 2024  
    •  
  • [C3] Empowering Personalized Learning through a Conversation-based Tutoring System with Student Modeling [paper]
      Minju Park, Sojung Kim, Seunghyun Lee, Soonwoo Kwon, Kyuseok Kim, CHI Late-Breaking Work 2024  
    •  
  • [C2] Addressing Selection Bias in Computerized Adaptive Testing: A User-Wise Aggregate Influence Function Approach [paper]
      Soonwoo Kwon*, Sojung Kim*, Seunghyun Lee, Jin-Young Kim, Suyeong An, Kyuseok Kim, CIKM 2023  
    •  
  • [C1] How Older Adults Use Online Videos for Learning [paper]
      Seoyoung Kim, Donghoon Shin, Jeongyeon Kim, Soonwoo Kwon, Juho Kim, CHI 2023  

Technical Reports

  • [T2] Marengo 2.7: Multi-Vector Representation for Multi-Modal Video Understanding [paper]
      TwelveLabs AI Team(Core), Tech Report 2025  
    •  
  • [T1] TWLV-I: Analysis and Insights from Holistic Evaluation on Video Foundation Models [paper]
      TwelveLabs AI Team, Tech Report 2024  

Education