Song Park
Song Park

Research Scientist

About Me

I am a Research Scientist, interested in understanding how deep neural networks perceive and process diverse visual concepts to enhance structured visual representations for real-world applications. I received my Ph.D. from Yonsei University in 2022 under the supervision of Hyunjung Shim. Following my doctoral studies, I worked as a Research Scientist at NAVER AI Lab from 2022 to 2025.

Download CV
Interests
  • Text-to-Image Generative Models
  • Visual Representation Learning
Education
  • Ph.D. and M.S. in Integrated Technology

    Yonsei University

  • B.S. in Integrated Technology

    Yonsei University

Research Interests

I am interested in understanding how deep neural networks (DNNs) perceive, represent, and process diverse visual concepts—such as mood, emotion, style, and semantics—and their impact on decision-making. My research aims to uncover the underlying mechanisms behind these representations to develop more structured and expressive visual features.

By enhancing the interpretability and robustness of learned representations, I seek to improve performance in real-world downstream tasks, including scene understanding, affective computing, and content generation.

Recent Publications
(2025). DNNs May Determine Major Properties of Their Outputs Early, with Timing Possibly Driven by Bias. arXiv preprint arXiv:2502.08167.
(2025). Probabilistic Language-Image Pre-Training. International Conference on Representation Learning.
(2024). Rotary position embedding for vision transformer. European Conference on Computer Vision.
(2024). SeiT++: Masked Token Modeling Improves Storage-efficient Training. European Conference on Computer Vision.
(2024). Similarity of neural architectures using adversarial attack transferability. European Conference on Computer Vision.