avatar

Seil Kang

PhD Student
Yonsei University
seil [at] yonsei [dot] ac [dot] kr

CV | Google Scholar | Github | Linkedin

Short Bio

I am Ph.D. student in Computer Science at Yonsei University.

My research focuses on investigating the behavior and inner workings of large-scale multimodal transformers, with an emphasis on interpretability-driven model improvement and alignment in systems such as Large Vision-Language Models (LVLMs) and Diffusion Transformers (DiTs).

Furthermore, I am also interested in research that analyzes and develops novel user experiences through the latest multimodal transformer science and engineering.

Last Update:

News

Publications [Google Scholar]

  1. Rare Text Semantics Were Always There in Your Diffusion Transformer teaser NeurIPS 2025
    Rare Text Semantics Were Always There in Your Diffusion Transformer
    Seil Kang*, Woojung Han*, Dayun Ju, Seong Jae Hwang
    NeurIPS 2025
    pdf

  2. Interpreting Attention Heads for Image-to-Text Information Flow in Large Vision-Language Models teaser NeurIPS 2025 W
    Interpreting Attention Heads for Image-to-Text Information Flow in Large Vision-Language Models
    Jinyeong Kim, Seil Kang, Jiwoo Park, Junhyeok Kim, Seong Jae Hwang
    NeurIPS 2025 Mechanistic Interpretability Workshop (Spotlight, <13%)
    pdf

  3. Neuron-Level Approach for Multi-Hop Reasoning in Large Vision-Language Models teaser Technical Report
    Neuron-Level Approach for Multi-Hop Reasoning in Large Vision-Language Models
    Seil Kang, Jinyeong Kim, Seong Jae Hwang
    Technical Report
    soon

  4. Your Large Vision Language Model Only Needs A Few Attention Heads for Visual Grounding teaser CVPR 2025
    Your Large Vision Language Model Only Needs A Few Attention Heads for Visual Grounding
    Seil Kang, Jinyeong Kim, Junhyeok Kim, Seong Jae Hwang
    CVPR 2025 (Highlight, <3%)
    pdf

  5. See What You Are Told: Visual Attention Sink in Large Multimodal Models teaser ICLR 2025
    See What You Are Told: Visual Attention Sink in Large Multimodal Models
    Seil Kang*, Jinyeong Kim*, Junhyeok Kim, Seong Jae Hwang
    ICLR 2025
    pdf

  6. FALCON: Frequency Adjoint Link with CONtinuous Density Mask for Fast Single Image Dehazing teaser CVPRW 2025
    FALCON: Frequency Adjoint Link with CONtinuous Density Mask for Fast Single Image Dehazing
    Donghyun Kim, Seil Kang, Seong Jae Hwang
    CVPRW 2025
    pdf

  7. WoLF: Wide-scope Large Language Model Framework for CXR Understanding teaser Technical Report
    WoLF: Wide-scope Large Language Model Framework for CXR Understanding
    Seil Kang, Donghyun Kim, Junhyeok Kim, Hyo Kyoung Lee, Seong Jae Hwang
    Technical Report
    pdf

  8. CoBra: Complementary Branch Fusing Class and Semantic Knowledge for Robust Weakly Supervised Semantic Segmentation teaser PR 2024
    CoBra: Complementary Branch Fusing Class and Semantic Knowledge for Robust Weakly Supervised Semantic Segmentation
    Woojung Han, Seil Kang, Kyobin Choo, Seong Jae Hwang
    PR 2024
    pdf