I am a Ph.D. candidate in Robotics at the Korea Advanced Institute of Science and Technology (KAIST), advised by Prof. Kuk-Jin Yoon in the Visual Intelligence Lab.
My research focuses on building robust and data-efficient omni-modal models with minimal human supervision.

I am open to research collaborations, coffee chats, and internship or full-time opportunities. Please feel free to contact me 🤗.

📝 Research Interests 

   🎯💡Keywords: Reliable real-world perception → Robust and data-efficient multimodal alignment → Reliable multimodal intelligence for agentic/physical AI

• My research aims to build reliable and data-efficient multimodal intelligence under practical constraints on data, compute, supervision, and sensing. I first encountered these challenges while developing real-time perception systems for autonomous driving under adverse weather using visual, LiDAR, and fused sensor inputs [J1, C1, C2, C5]. This work showed that reliable deployment requires models to generalize beyond controlled training conditions while satisfying strict requirements on latency, safety, and computational efficiency.

• I subsequently investigated model reliability more directly by developing transferable adversarial methods that expose vulnerabilities in black-box systems across different architectures [W2→C3, C4, W3→C6]. These studies showed that reliability depends not only on the robustness of individual encoders, but also on whether their learned representations preserve meaningful structure when models, inputs, or environments change. This perspective led me to vision-language learning, where the central bottleneck is often not model scale alone, but the cost, noise, and limited correspondence of multimodal supervision. My current work therefore develops data-efficient and geometry-aware methods for identifying and preserving useful cross-modal correspondence under limited and imperfect image-text supervision [C7, C8].

• Building on this foundation, my future research comprises three connected directions: data-efficient multimodal learning, robust cross-modal alignment, and multi-modal models for agentic/physical AI. I aim to develop compact and transferable supervision for training and adapting multimodal models across architectures, modalities, and downstream tasks. I then seek to align independently pretrained modality experts so that they can reinforce one another without retraining from scratch, while remaining reliable under distribution shifts, corrupted or missing modalities, and adversarial perturbations. In the future, I plan to connect multimodal representations with perception, prediction, memory, and action, enabling embodied systems and world models to learn efficiently from heterogeneous real-world experience. My long-term goal is to establish general principles and open evaluation frameworks for omnimodal models that are efficient, robust, and deployable beyond controlled benchmarks.

📌 Education

• Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea

     🏫 Ph.D. in Robotics, Advisor: Prof. Kuk-Jin Yoon, 2022.08 - Present

     🎓 M.S. in Electrical Engineering, Advisor: Prof. Jong-Hwan Kim, 2020.08 - 2022.08

• The Cooper Union for the Advancement of Science and Art, New York, NY, USA

     🎓 B.E. in Electrical Engineering, 2014.09 - 2020.05 (*2016-2018: military duty leave)

📰 News

10  '26 — Selected as a Top Reviewer @ NeurIPS 2026 (Main conference)
09 '26 — 1 paper accepted at ECCV 2026 and also presented at the CDEL Workshop.
06 '26 — 1 paper presented at CVPR 2026 and also at the DataMFM Workshop.
04 '26 — 1 paper accepted at ICLR 2026.
03 '26 — 1 paper accepted at WACV 2026.

📖 Publications    (W: Peer-reviewed Workshop, C: Peer-reviewed Conference, J: Peer-reviewed Journal, P: Pre-print, K-C/W: Korean Conference/Workshop)
      Research Areas:

• Computer Vision (Robust/data-efficient representation learning, Multimodal learning) [P1, P3, C5, C7, C8] 

• Machine Learning (Robust, safe, trustworthy, adversarial AI) [J2, W2→C3, C4, W3→C6]

  • Robot/Machine Vision (Autonomous driving, sensor fusion) [J1, C1, K-C1, C2, P2]

🗂️ Research Projects

• C-arm X-ray real-time denoising engine development (DK Medical Systems, Jul.'25 – Dec.'26)

• AI/SW Project for Commissioned Education (Ministry of National Defense (MND), Jun.'22 – Dec.'26)

• Synthetic Military Target Data Generation for Domain Adaptive Object Detection (LIG Defense&Aerospace(Nex1), Oct.’22 – Sep.’23)

• Development of Artificial Intelligence Technology that Continuously Improves Itself as the Situation Changes in the Real World (Korea MSIT (IITP), No.2020-0-00440, Sep.’20 –Aug.’22)

• Development of Real-time Smart Solution for Solder Mount Technology (SMT) (Koh-Young Tech., Sep.’20 – Aug.’22)

📤 Academic Activities

Reviewer

• AAAI, ICLR, NeurIPS, 2026 - Present

• IEEE Transactions on Information Forensics and Security (T-IFS), 2026 - Present

  • IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT), 2026 - Present

  • IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2026 - Present

  • IEEE Transactions on Multimedia (T-MM), 2026 - Present

• IEEE Transactions on Cybernetics (T-CYB), 2026 - Present

• IEEE Transactions on Geoscience and Remote Sensing (T-GRS), 2024 - Present

  • IEEE Transactions on Intelligent Transportation Systems (T-ITS), 2022 - Present

• IEEE Access, 2022 - Present

• NeurIPS Workshop on Machine Learning for Health (ML4H), 2020 - Present

Professional Society Membership

• • IEEE Computer Society, 2025

  • Order of the Engineer, 2020 – Present

  • IEEE-Eta Kappa Nu (Student Member) 2018 – Present

  • ACM (Student Member) 2018 – Present