Seongheon Park

Contact. seongheon_park [at] cs [dot] wisc [dot] edu

prof_pic.jpg

1210 W Dayton St

Madison, WI 53706

Hello! I’m a second-year PhD student in the Computer Sciences department at the University of Wisconsin-Madison, where I am fortunate to be advised by Prof. Sharon Yixuan Li. Previously, I completed my MS degree at Yonsei University in the Electrical and Electronic Engineering department under the supervision of Prof. Kwanghoon Sohn and Prof. Kibok Lee.

My research interests lie broadly in safe and reliable foundation models, with a current focus on large language models, multi-modal language models, and diffusion models. By “reliable,” I mean developing methods that help users better trust model outputs. Specifically, my work centers on:

  • Detecting and mitigating hallucinations and errors
  • Uncertainty quantification
  • Interpretability

I am also interested in applying these techniques to broader challenges, including the design of more accurate reward and training signals for post-training and self-verification/improvement.

:mega: I am actively seeking internship opportunities for Summer 2026. Please feel free to reach out if there is a potential fit!

news

Sep 18, 2025 Happy to share that two of our papers have been accepted at NeurIPS 2025 :blossom:
Sep 04, 2025 Our paper HalluEntity: Benchmarking and Understanding Entity-Level Hallucination Detection is accepted by TMLR :blush:
Jul 02, 2025 Our paper about LVLM object hallucnation detection is accepted by ICML 2025 Workshop on Reliable and Responsible Foundation Models :notes:
May 01, 2025 Two papers are accepted by ICML 2025 :fire:
Apr 02, 2025 Two papers are accepted by ICLR 2025 Workshop: Quantify Uncertainty and Hallucination in Foundation Models :relaxed:

publications

  1. Preprint
    lp.png
    Understanding Language Prior of LVLMs by Contrasting Chain-of-Embedding
    Lin Long, Changdae Oh, Seongheon Park, and Yixuan Li
    arXiv:2509.23050, Sep 2025
  2. Preprint
    shaking_preview.png
    Shaking to Reveal: Perturbation-Based Detection of LLM Hallucinations
    Jinyuan Luo, Zhen Fang, Yixuan Li, Seongheon Park, and Ling Chen
    arXiv:2506.02696, Jun 2025
  3. NeurIPS
    glsim_preview.png
    GLSim: Detecting Object Hallucinations in LVLMs via Global-Local Similarity
    Seongheon Park, and Yixuan Li
    Conference on Neural Information Processing Systems, Jul 2025
  4. NeurIPS
    georanker_preview.png
    GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization
    Pengyue Jia, Seongheon Park, Song Gao, Xiangyu Zhao, and Yixuan Li
    Conference on Neural Information Processing Systems, May 2025
  5. ICML
    tsv_preview.png
    Steer LLM Latents for Hallucination Detection
    Seongheon Park, Xuefeng Du, Min-Hsuan Yeh, and Yixuan Li
    International Conference on Machine Learning, May 2025
  6. ICML
    position_preview.png
    Position: Challenges and Future Directions of Data-Centric AI Alignment
    Min-Hsuan Yeh, Jeffrey Wang, Xuefeng Du, Seongheon Park, Leitian Tao, Shawn Im, and Yixuan Li
    International Conference on Machine Learning, May 2025
  7. TMLR
    can_preview.png
    HalluEntity: Benchmarking and Understanding Entity-Level Hallucination Detection
    Min-Hsuan Yeh, Max Kamachee, Seongheon Park, and Yixuan Li
    Transactions on Machine Learning Research, Mar 2025
  8. CVPR Workshop
    owssl_preview.png
    Rethinking Open-World Semi-Supervised Learning: Distribution Mismatch and Inductive Inference
    Seongheon Park*, Hyuk Kwon*, Kwanghoon Sohn, and Kibok Lee
    CVPR Workshop on Computer Vision in the Wild, Jun 2024
  9. ICCV
    czsl_preview.png
    Hierarchical visual primitive experts for compositional zero-shot learning
    Hanjae Kim, Jiyoung Lee, Seongheon Park, and Kwanghoon Sohn
    IEEE/CVF International Conference on Computer Vision, Oct 2023
  10. CVPR
    partmix_preview.png
    Partmix: Regularization strategy to learn part discovery for visible-infrared person re-identification
    Minsu Kim, Seungryong Kim, Jungin Park, Seongheon Park, and Kwanghoon Sohn
    IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun 2023
  11. WACV
    ngmil_preview.png
    Normality guided multiple instance learning for weakly supervised video anomaly detection
    Seongheon Park, Hanjae Kim, Minsu Kim, Dahye Kim, and Kwanghoon Sohn
    IEEE/CVF Winter Conference on Applications of Computer Vision, Jan 2023
  12. WACV
    zsvg_preview.png
    Language-free training for zero-shot video grounding
    Dahye Kim, Jungin Park, Jiyoung Lee, Seongheon Park, and Kwanghoon Sohn
    IEEE/CVF Winter Conference on Applications of Computer Vision, Jan 2023