I am a Research Officer at the Agency for Defense Development,
Korea's national defense R&D agency (similar to DARPA), and a First Lieutenant in the Republic of Korea Army.
I received my B.S. in Computing with a minor in Physics from KAIST.
My research aims to build cost-efficient and reliably evaluated generative AI systems
by uncovering the mathematical structure beneath the heuristics and metrics the field relies on.
Publications
ICML Workshop CTB 2026
Rethinking FID Through the Geometry of the Reference Dataset
Yunghee Lee, Byeonghyun Pak
Reference dataset geometry significantly moderates FID, so distributional metrics should be reported alongside dataset geometry.
CVPR Workshop PVUW 2026
Pixel-level Scene Understanding in One Token: Visual States Need What-is-Where Composition
Seokmin Lee, Yunghee Lee, Byeonghyun Pak, Byeongju Woo
CroBo learns compact visual states that capture what-is-where pixel-level scene composition, enabling stronger what-moves-where dynamic scene understanding.
CVPR Workshop ECV 2024
Oral Presentation
LVS: A Learned Video Storage for Fast and Efficient Video Understanding
Yunghee Lee, Jongse Park
LVS memoizes feature vectors for already-seen video clips and reuses them for future video understanding queries.
IEEE Robotics Autom. Lett. (RA-L) 2023
HybGrasp: A Hybrid Learning-to-Adapt Architecture for Efficient Robot Grasping
Jungwook Mun, Khang Truong Giang, Yunghee Lee, Nayoung Oh, Sejoon Huh, Min Kim, Sungho Jo
HybGrasp efficiently learns to adapt to different gripper designs by refining the base pose with lightweight modules.
Experience
Independent Research
Research Collaboration with Harvard University (with Jaeyeon Kim)
- Decomposed the Fréchet distance exactly over interpretable feature directions. [Under review at ICLR 2027]
Agency for Defense Development (ADD)
Research Officer, Manager: Dr. Eunjin Koh, Advisor: Dr. Hoseong Kim
- Accelerated diffusion-based synthetic data generation by ~30% with multirate integration (THG). [NeurIPS 2025]
- Found FID and ImageReward disagreeing while evaluating THG, which led to a study of FID. [ICMLW 2026]
- Rebuilt the data collection software for surveillance UAVs, enabling longer flights.
Computer Architecture and Systems Lab @ KAIST
Undergraduate Research Intern, Advisor: Prof. Jongse Park
- Worked on efficient video understanding systems. [CVPRW 2024]
Neuro-Machine Augmented Intelligence Lab @ KAIST
Undergraduate Research Intern, Advisor: Prof. Sungho Jo
- Worked on robot grasping for various gripper designs. [RA-L 2023]
Education
Korea Advanced Institute of Science and Technology (KAIST)
B.S. in Computing, Summa Cum Laude