Taeheon Kim
Ph.D. Candidate
Integrated Vision and Language Lab
Korea Advanced Institute of Science and Technology (KAIST)
e-mail: eetaekim@kaist.ac.kr [Google Scholar] [Youtube]
Embedded Files
I'm a final year Ph.D. student at KAIST, advised by Professor Yong Man Ro.
My Ph.D. research focuses on understanding failures of AI perception and reasoning in real-world physical conditions, and on developing robust multi-modal learning methods to address such failures, with applications to autonomous and safety-critical systems.
Research Interest
Multimodal AI under Physical-world Conditions
Robust and Failure-Aware Multi-modal Learning
Trustworthy Multimodal AI Systems
Education
Education
KAIST, Daejeon, South Korea
Ph.D. in Electrical Engineering, Mar. 2020 – Present
KAIST, Daejeon, South Korea
B.S. in Electrical Engineering, Mar. 2014 – Feb. 2019
Georgia Institute of Technology, Atlanta, GA, USA
Exchange Student, Computer Engineering, Aug. 2016 – May 2017
GPA: 4.0 / 4.0, Faculty Honors
Gyeongnam Science High School, South Korea
Early Graduation (2-year Program), Mar. 2012 – Feb. 2013
Publications
Research Program: Failure Discovery → Physical Realization → Causal Analysis → Failure Diagnosis → Failure Correction (Feature / Reasoning)
This program investigates how and why AI perception fails under real-world physical conditions, and how such failures can be systematically analyzed and corrected.
The publications below are listed in reverse chronological order and collectively realize
the above research program.
[Failure Diagnosis & Structural Removal]
7. Infrared Cleanser: Detecting the Hyperchannel and Removing Adversarial Patches in Infrared Images
Failure diagnosis and channel-level removal of physical-world vulnerabilities in infrared perception
Taeheon Kim, Sangyun Chung, Youngjoon Yu, and Yong Man Ro
Under Review (2024)
[Failure Correction – Reasoning Level]
6. MSCoTDet: Language-driven Multi-modal Fusion for Improved Multispectral Pedestrian Detection
Failure correction via language-driven reasoning for modality bias in multispectral perception
Taeheon Kim*, Sangyun Chung*, Damin Yeom, Youngjoon Yu, Hak Gu Kim, and Yong Man Ro (* : equally contributed.)
IEEE Transactions on Circuits and Systems for Video Technology (TCSVT), 2025
Impact Factor: 11.1, JCR Top 3.7%
[paper]
5. Revisiting Misalignment in Multispectral Pedestrian Detection: A Language-driven Approach for Cross-modal Alignment Fusion
Analysis of cross-modal misalignment causing perception failures under physical conditions
Taeheon Kim*, Sangyun Chung*, Youngjoon Yu*, and Yong Man Ro (* : equally contributed.)
IEEE International Conference on Image Processing (ICIP), 2024
[Causal Analysis]
4. Causal Mode Multiplexer: A Novel Framework for Unbiased Multispectral Pedestrian Detection
Causal analysis of modality bias causing perception failures under real-world physical conditions
Taeheon Kim*, Sebin Shin*, Youngjoon Yu, Hak Gu Kim, and Yong Man Ro (* : equally contributed.)
IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR), 2024
Acceptance Rate 23.6%
[paper][demo video][code]
[Physical Realization]
3. Multispectral Invisible Coating: Laminated Visible-Thermal Physical Attack against Multispectral Object Detectors using Transparent Low-e films
Physical realization and amplification of multispectral perception failures
Taeheon Kim, Youngjoon Yu, and Yong Man Ro
AAAI Conference on Artificial Intelligence (AAAI), 2023
Acceptance Rate 19.6%
[paper][demo video]
[Failure Correction – Feature Level]
2. Defending Physical Adversarial Attack on Object Detection via Adversarial Patch-Feature Energy
Failure correction via feature-level robustness against physical adversarial conditions
Taeheon Kim, Youngjoon Yu, and Yong Man Ro
ACM Multimedia Conference (MM), 2022
Acceptance Rate 24%
[paper][demo video]
[Failure Discovery]
1. MAP: Multispectral Adversarial Patch to Attack Multispectral Person Detection
Failure discovery under physical adversarial conditions in multispectral perception
Taeheon Kim, Hong Joo Lee, and Yong Man Ro
IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2022
Professional Activities
■ Reviewer
IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2024, 2025, 2026
IEEE / CVF International Conference on Computer Vision (ICCV) 2025
AAAI Conference on Artificial Intelligence (AAAI) 2025, 2026
International Conference on Learning Representations (ICLR) 2025
IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) 2023, 2024
IEEE Transactions on Neural Networks and Learning Systems (TNNLS) 2024
IEEE Transactions on Image Processing (TIP) 2025
IEEE Transactions on Multimedia (TMM) 2025
IEEE Transactions on Intelligent Transportation Systems (T-ITS) 2025
IEEE Robotics and Automation Letters (RA-L) 2025
IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2023
IEEE International Conference on Image Processing Workshop (ICIPW) 2024
■ Teaching
2016 Spring: CS101: Introduction to Programming
@KAIST, Undergraduate Teaching Assistant
2017 Spring: ECE2031: Digital Design Laboratory
@Georgia Tech, Undergraduate Teaching Assistant
Supervised over 30 Georgia Tech students in laboratory experiments, technical writing, and presentation skills.
2021 Spring: EE202: Signals and Systems
@KAIST, Graduate Teaching Assistant
2021 Fall, 2022 Spring, 2022 Fall: EE305: Introduction to Electronics Design Lab
@KAIST, Graduate Teaching Assistant
Other Skills
Classical piano performance, with a focus on Chopin’s works. Chopin's music serves as a source of creativity and inspiration for my research.
Classical Piano Performances (Played by Taeheon Kim)
Chopin's Etudes (Op.10 No.1, Op.25 No.2, Op.25 No.9, Op.25 No.12)
Chopin's Piano Sonata (No.3, Op.58 4th mov)
Page updated
Google Sites
Report abuse