Hi! I am a PhD student in AI at Korea University, advised by Donghyun Kim. I build multimodal agents that find and verify visual evidence in 3D scenes and long videos. Before my PhD, I spent 3 years as an AI Research Engineer at VUNO Inc., completing my military service as Technical Research Personnel.
I am looking for research internships in 2027.
Research
MLLMs cannot attend to every region or frame they are shown, so they must locate the few pieces of evidence a question depends on. I study how they find, use, and verify that evidence across space and time:
- Spatial: 3D spatial reasoning with visual tools as scaffolding for RL exploration (ICML'26) and spatially grounded robot planning (IROS'26), toward agents that act in the physical world.
- Temporal: narrative-level retrieval units for hours-long videos (NeurIPS'26 E&D) and grounding confusable movie moments (Preprint'26), toward agents that verify evidence in long videos.
- Spatiotemporal (current focus): bringing the two together for embodied perception, where an agent moving through a scene must tell its own motion apart from that of objects and keep track of how the layout changes, with answers that truly rest on what it saw.
Earlier work: medical imaging, image segmentation (ICCV-W'25, Preprint'24), and self-supervised / domain-adaptive learning (ACCV'24).
News
- 10/2026: Released our preprint "Grounding Confusable Movie Moments" on arXiv.
- 09/2026: "Beyond Visual Boundaries: Rethinking Scene Segmentation for Movie RAG" is accepted to the NeurIPS 2026 Evaluations and Datasets Track.
- 06/2026: "Spatially Grounded Long-Horizon Task Planning in the Wild" is accepted to IROS 2026.
- 05/2026: "Diversity Over Frequency: Rethinking Tool Use in Visual Chain-of-Thought Agents" is accepted to ICML 2026.
- 11/2025: Received the Seoul AI Tech Graduate Student Scholarship.
- 07/2025: "SynRES" is accepted to the ICCV 2025 DataCV Workshop.
- 06/2025: Selected for the SKT Devocean AI Fellowship.
- 03/2025: Started my PhD in Artificial Intelligence at Korea University.
- 11/2024: Presented at RSNA 2024 and received the Best Abstract Award (Oral) at KCR 2024.
- 09/2024: "CNG-SFDA" is accepted to ACCV 2024.
- 07/2024: Released the "Mask-JEPA" preprint.
- 06/2022: Won the TIGER Challenge (Leaderboard 2).
Selected Publications
Work Experience
Developed deep-learning models for medical imaging across the BRAIN and Digital Oncology teams, working on brain-MRI parcellation and report generation, clinical validation of DeepBrain products, and cell/lesion segmentation for digital pathology (TIGER Challenge winner).
Education
Awards
Services
ICML 2026 (Gold Reviewer), NeurIPS 2026, BMVC 2026, AAAI 2026, ICLR 2027
Miscellaneous
Outside of research, I enjoy escape rooms, so feel free to reach out if you want to team up! I am also broadly interested in applying AI across a wide range of domains.