Changyeon Kim

changyeon.kim [AT] kaist.ac.kr

changyeon.jpg

Hello. I am a Ph.D. student at KAIST, advised by Kimin Lee and Jinwoo Shin, and a research scientist intern at NVIDIA Seattle Robotics Lab working with Yijie Guo and Yashraj Narang. During my Ph.D., I was a visiting scholar at UT Austin advised by Yuke Zhu

My research aims to enable robotic agents to continually improve through real-world deployment with reduced human effort. To this end, I develop scalable and reliable methods across the real-world robot post-training pipeline, including offline reinforcement learning (RL) for large vision-language-action (VLA) policies dealing with long action chunks, rollout-free validation for reliable checkpoint selection during training, and lifelong adaptation that improves learning efficiency while mitigating catastrophic forgetting. Previously, my work focused on scalable reward learning methods that capture human intent by leveraging human feedback and vision-language foundation models.

Prior to my graduate studies, I was a machine learning engineer at Recommendation Team of Kakao. Before that, I completed my BS in Computer Science at KAIST.


News

May 13, 2026 I received Silver Reviewer Award in ICML 2026 :smile:.
May 4, 2026 I joined NVIDIA Seattle Robotics Lab as a research scientist intern!
Apr 30, 2026 Two papers (forget-me-not and RS-CL) are accepted to ICML 2026. :smile:.
Especially, forget-me-not is accepted as oral (top 0.7%).
See you in Seoul 🇰🇷!
Feb 26, 2026 XVR is accepted to CVPR 2026. :smile:.
Jan 26, 2026 Three papers (DEAS, HAMLET, and MG-Select) are accepted to ICLR 2026. :smile:.
DEAS is my first-authored paper from my UT Austin visit.
See you in Rio 🇧🇷!

Publications

  1. ICML
    Pretrained Vision-Language-Action Models are Surprisingly Resistant to Forgetting in Continual Learning
    Huihan Liu,  Changyeon Kim, Bo Liu, Minghuan Liu, and Yuke Zhu
    In International Conference on Machine Learning (ICML), 2026
    Oral presentation (168 / 23918 = 0.7%)
    arXiv Code Website
  2. ICML
    Contrastive Representation Regularization for Vision-Language-Action Models
    Taeyoung Kim, Jimin Lee, Myungkyu Koo, Dongyoung Kim, Kyungmin Lee,  Changyeon Kim, Younggyo Seo, and Jinwoo Shin
    In International Conference on Machine Learning (ICML), 2026
    arXiv
  3. CVPR
    Learning Multi-View Spatial Reasoning from Cross-View Relations
    Suchae Jeong*, Jaehwi Song*, Haeone Lee, Hanna Kim, Jian Kim, Dongjun Lee, Dong Kyu Shin,  Changyeon Kim, Dongyoon Hahm, Woogyeol Jin, Juheon Choi, and Kimin Lee
    In Conference on Computer Vision and Pattern Recognition (CVPR), 2026
    (*: equal contribution)
  4. ICLR
    DEAS: DEtached value learning with Action Sequence for Scalable Offline RL
    Changyeon Kim, Haeone Lee, Younggyo Seo, Kimin Lee^, and Yuke Zhu^
    In International Conference on Learning Representations (ICLR), 2026
    (^: equal advising)
    arXiv PDF Website
  5. ICLR
    Verifier-free Test-Time Sampling for Vision Language Action Models
    Suhyeok Jang, Dongyoung Kim,  Changyeon Kim, Youngsuk Kim, and Jinwoo Shin
    In International Conference on Learning Representations (ICLR), 2026
    arXiv
  6. ICLR
    HAMLET: Switch Your Vision-Language-Action Model into a History-Aware Policy
    Myungkyu Koo, Daewon Choi, Taeyoung Kim, Kyungmin Lee,  Changyeon Kim, Younggyo Seo, and Jinwoo Shin
    In International Conference on Learning Representations (ICLR), 2026
    arXiv Website
  7. CoLLAs
    B-MoCA: Benchmarking Mobile Device Control Agents across Diverse Configurations
    Juyong Lee, Taywon Min, Minyong An, Dongyoon Hahm, Haeone Lee,  Changyeon Kim, and Kimin Lee
    In Conference on Lifelong Learning Agents (CoLLAs), 2025
    Previously accepted to ICLR 2024 Workshop on Generative Models for Decision Making as a spotlight presentation
    arXiv Code Website
  8. ICLR
    Subtask-Aware Visual Reward Learning from Segmented Demonstrations
    Changyeon Kim, Minho Heo, Doohyun Lee, Jinwoo Shin, Honglak Lee, Joseph J. Lim^, and Kimin Lee^
    In International Conference on Learning Representations (ICLR), 2025
    (^: equal advising)
    arXiv PDF Code Website
  9. NeurIPS
    Guide Your Agent with Adaptive Multimodal Rewards
    Changyeon Kim, Younggyo Seo, Hao Liu, Lisa Lee, Jinwoo Shin, Honglak Lee, and Kimin Lee
    In Conference on Neural Information Processing Systems (NeurIPS), 2023
    Previously accepted to ICML 2023 Workshop on New Frontiers in Learning, Control, and Dynamical Systems
    Finalist of Qualcomm Innovation Fellowship 2024 Korea
    arXiv PDF Code Website
  10. ICLR
    Preference Transformer: Modeling Human Preferences using Transformers for RL
    Changyeon Kim*, Jongjin Park*, Jinwoo Shin, Honglak Lee, Pieter Abbeel, and Kimin Lee
    In International Conference on Learning Representations (ICLR), 2023
    (*: equal contribution)
    arXiv PDF Code Website
  11. NeurIPSW
    Dynamics-Augmented Decision Transformer for Offline Dynamics Generalization
    Changyeon Kim*, Junsu Kim*, Younggyo Seo, Kimin Lee, Honglak Lee, and Jinwoo Shin
    In Conference on Neural Information Processing Systems (NeurIPS) Workshop on Offline Reinforcement Learning, 2022
    (*: equal contribution)
    PDF

Work Experience


NVIDIA Seattle Robotics Lab
Research Intern (May 2026 ~ Jul 2026) (Expected)

The University of Texas at Austin
Visiting scholar working with Prof. Yuke Zhu (July 2024 - Jun 2025)

Recommendation Team, Kakao
Machine Learning Engineer (Dec 2020 ~ Feb 2022)

Data Science Group, Institute of Basic Science
Resarch Intern advised by Prof. Meeyoung Cha (Jul 2019 - Nov 2020)


Honors and Awards

Silver Reviwer Award,  International Conference on Machine Learning (ICML), 2026
Notable Reviewer,  International Conference on Learning Representations (ICLR), 2025
Finalist,  Qualcomm Innovation Fellowship 2024 Korea
Travel Award ($2,000),  Conference on Neural Information Processing Systems (NeurIPS), 2023
Recipient ($3,000),  KAIST-Google Partnership Program, 2023
Recipient ($2,000),  Google East Asia Student Travel Grant, 2023
Travel Award ($1,000),  International Conference on Learning Representations (ICLR), 2023
Dean's List,  KAIST Department of Engineering, 2019
Recipient ($5,000),  Line Scholarship, 2019
Recipient,  National Science and Engineering Scholarship, Korea Ministry of Science and ICT, 2017 - 2019
Recipient ($3,000),  Kwanjeong Scholarship, 2017


Invited Talks

Guide Your Agent with Adaptive Multimodal Rewards
LG AI Research (New Orleans, LA, USA)


Academic Services

Conference Reviewer:  ICLR, ICML, NeurIPS, CoRL, AAAI, RSS, ECCV, RA-L
Workshop Reviewer:  Frontiers4LCD@ICML'23, MRM-D@CoRL'24, WRM@NeurIPS'26