Heekyung (Anne) Lee

prof_pic_2.JPG

Hi! I am Heekyung(Anne) Lee. I am an Undergraduate Student studying Computer Science at POSTECH, South Korea. (Research Advisor: Prof. Minsu Cho) I am also an undergraduate researcher at Berkeley AI Research (BAIR), fortunate to be supervised under Prof. Trevor Darrell’s group. (Advisors: Prof. Trevor Darrell, Prof. Sewon Min, and David M. Chan)

I am interested in building trustworthy multi-modal models, that can flexibly adapt to the user’s needs, acting naturally and efficiently in complex multi-modality environments.

I am recently excited about the two questions below:

  • Controllable and Adaptable Vision-Language Models
    ∘ How can we build controllable and adaptable intelligence that can adapt their behavior to the user's needs?
    ∘ What is Hallucination in multi-modal models, and what is the best recipe to mitigate it?
  • Real-World Multimodal Understanding
    ∘ How can we represent the world as input data that reflects the world's complexity in a better way?
    ∘ How can we build models that are capable of handling complex multi-modality environments as an input? (e.g. how should models know which modality to focus on for a given task?)

Feel free to check out my publications and projects.

.

And also, I am a huge sports person! I generally enjoy playing soccer and running. Let’s be Strava friends! (Follow me here)

news

Oct 21, 2025 🔥 Check out our new paper Constantly Improving Image Models Need Constantly Improving Benchmarks! Link to 🧵: @ECHO.
Sep 19, 2025 ✈️ I am attending NeurIPS 2025 with our work on Generate, but Verify: Reducing Visual Hallucination in Vision-Language Models with Retrospective Resampling. Very excited to see you all in person!
Aug 20, 2025 ✈️ I am attending EMNLP 2025 to present my work on Puzzled by Puzzles: When Vision-Language Models Can’t Take a Hint. See you in Suzhou, China!
Aug 14, 2025 🇩🇪 I obtained A1 level certificate in Goethe-Zertifikat in German language!

publications

  1. reverse.png
    REVERSE: Reducing Hallucination in Vision-Language Models with Retrospective Resampling
    Tsung-Han Wu, Heekyung Lee, Jiaxin Ge, Joseph E. Gonzalez, Trevor Darrell, and David M. Chan
    2025
    In Thirty-Ninth Conference on Neural Information Processing Systems (NeurIPS), 2025
  2. visual-puzzles.png
    Puzzled by Puzzles: When Vision-Language Models Can’t Take a Hint
    Heekyung Lee, Jiaxin Ge, Tsung-Han Wu, Minwoo Kang, Trevor Darrell, and David M. Chan
    2025
    In 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2025
  3. echo.png
    Constantly Improving Image Models Need Constantly Improving Benchmarks
    Jiaxin Ge*, Grace Luo*, Heekyung Lee, Nishant Malpani, Long Lian, XuDong Wang, Aleksander Holynski, Trevor Darrell, Sewon Min, and David M. Chan
    2025
    arXiv preprint arXiv:2510.15021