Changdae Oh

Changdae Oh

Ph.D. student at Department of Computer Sciences, University of Wisconsin-Madison

I am a Ph.D. student in the Department of Computer Sciences at the University of Wisconsin–Madison, advised by Prof. Sharon Li, and an incoming intern at Meta Superintelligence Labs. Prior to my Ph.D., I earned my M.S. degree in AI from the University of Seoul, where I was advised by Prof. Kyungwoo Song and Prof. Jiyoung Jung. I also had the opportunity to work with Zhi-Qi Cheng, Alexander Hauptmann, and David Mortensen during a visiting period at Carnegie Mellon University, and with Dongyoon Han and Sangdoo Yun during my internship at NAVER AI Lab.

I am broadly interested in machine learning fundamentals and trustworthy AI. Recently, I have been focusing on understanding and improving the robustness of multimodal LLMs under distribution shifts and uncertainty quantification of LLM agents.

News

Feb. 2026, Will be interning for Meta Superintelligence Labs (PAR Team) this summer at Menlo Park, CA!
Jan. 2026, Start collaboration with Argonne National Laboratory as a Visiting Student-Subcontractor!
Jan. 2026, Three papers got accepted to ICLR 2026!

Selected Publications and Preprints

(* denotes equal contribution)
Refer to the Google Scholar and CV for the full publication list.

  • Understanding Language Prior of LVLMs by Contrasting Chain-of-Embedding
    Lin Long*, Changdae Oh*, Seongheon Park, Sharon Li
    [paper] [code]
    ICLR 2026

  • How Do Transformers Learn to Associate Tokens: Gradient Leading Terms Bring Mechanistic Interpretability
    Shawn Im, Changdae Oh, Zhen Fang, Sharon Li
    [paper] [code]
    ICLR 2026

  • General Exploratory Bonus for Optimistic Exploration in RLHF
    Wendi Li, Changdae Oh, Sharon Li
    [paper] [code]
    ICLR 2026
    NeurIPS 2025, Workshop on Socially Responsible and Trustworthy Foundation Models (Oral Presentation; 9/136=6.6%)

  • Visual Instruction Bottleneck Tuning
    Changdae Oh, Jiatong Li, Shawn Im, Sharon Li
    [paper] [code]
    NeurIPS 2025
    ICML 2025, Workshop on Reliable and Responsible Foundation Models (Oral Presentation; 6/176=3.4%)

  • Understanding Multimodal LLMs Under Distribution Shifts: An Information-Theoretic Approach
    Changdae Oh, Zhen Fang, Shawn Im, Xuefeng Du, Yixuan Li
    [paper] [code]
    ICML 2025
    ICLR 2025, QUESTION Workshop (Oral Presentation)

  • DaWin: Training-free Dynamic Weight Interpolation for Robust Adaptation
    [paper][code]
    Changdae Oh, Yixuan Li, Kyungwoo Song, Sangdoo Yun, Dongyoon Han
    ICLR 2025
    NeurIPS 2024, Workshop on Adaptive Foundation Models

  • Towards Calibrated Robust Fine-Tuning of Vision-Language Models
    Changdae Oh*, Hyesu Lim*, Mijoo Kim, Dongyoon Han, Sangdoo Yun, Jaegul Choo, Alexander Hauptmann, Zhi-Qi Cheng, Kyungwoo Song
    [paper] [code]
    NeurIPS 2024
    NeurIPS 2023, Workshop on Distribution Shifts

  • Geodesic Multi-Modal Mixup for Robust Fine-tuning
    Changdae Oh*, Junhyuk So*, YongTaek Lim, Hoyoon Byun, Minchul Shin, Jong-June Jeon, Kyungwoo Song
    [paper] [code]
    NeurIPS 2023

  • BlackVIP: Black-Box Visual Prompting for Robust Transfer Learning
    [paper] [code]
    Changdae Oh, Hyeji Hwang, Hee-young Lee, YongTaek Lim, Geunyoung Jung, Jiyoung Jung, Hosik Choi, Kyungwoo Song
    CVPR 2023

  • Learning Fair Representation via Distributional Contrastive Disentanglement
    [paper] [code]
    Changdae Oh, Heeji Won, Junhyuk So, Taero Kim, Yewon Kim, Hosik Choi, Kyungwoo Song
    KDD 2022

Education

Experience

  • Research Scientist Intern, Meta Superintelligence Labs
    Mentor: Julian Katz-Samuels, May. 2026 ~ current
  • Visiting Student-Subcontractor, Argonne National Laboratory
    Mentor: Tanwi Mallick, Feb. 2026 ~ current
  • Research Intern, NAVER AI Lab
    Mentor: Dongyoon Han and Sangdoo Yun, Apr. 2023 ~ Aug. 2024
    • DaWin: Training-free Dynamic Weight Interpolation for Robust Adaptation, ICLR 2025
  • Visiting Scholar, Carnegie Mellon University
    Mentor: Zhi-Qi Cheng, Sep. 2023 ~ Feb. 2024
    • Towards Calibrated Robust Fine-Tuning of Vision-Language Model, NeurIPS 2024
    • Mitigating the Linguistic Gap with Phonemic Representations for Robust Cross-lingual Transfer, EMNLP 2024 Workshop

Talks

  • Jan. 2026, MLAI Lab @ Yonsei University, “On the Dynamic Reliability of Adaptive Foundation Models”
  • Jun. 2025, ResearchTrend.AI, “Visual Instruction Bottleneck Tuning”

Academic Services