Ph.D. student in Artificial Intelligence · KAIST
I am a Ph.D. student at the Kim Jaechul Graduate School of AI, KAIST, advised by Prof. Jaesik Choi in the Statistical Artificial Intelligence Lab.
I study how neural networks represent concepts and use multimodal information. My research focuses on interpretability, including concept circuits and polysemantic features, and multimodal language models.
I have worked on multimodal model evaluation at NAVER Cloud and industrial time-series modeling at Gauss Labs and INEEJI. I am interested in postdoctoral and research opportunities.
* Equal contribution
Findings of EMNLP 2026 · Accepted
A model can preserve what it hears without using it to answer. Tracing acoustic information through the encoder, projector, and LLM reveals a readout bottleneck.
ECCV 2026
We introduce a general, scalable framework for disentangling polysemanticity through attribution-footprint clustering, without prescribing the number of concepts each neuron represents.
Also presented at the ICML 2026 Workshop on Mechanistic Interpretability.
ACM Computing Surveys 2026
This survey unifies time-series methods across tasks through a taxonomy of transparency and evaluates explanations by their granularity and mechanistic depth.
ICCV 2025
We discover circuits for fine-grained visual concepts by tracing functional dependencies and semantic alignment between neurons across layers.
ICPRAI 2024
Counterfactual examples guided by a correctness prober expose classifier vulnerabilities without requiring ground-truth labels for the inputs being analyzed.
NeurIPS 2023
Learning agreement need not be symmetric. Guided stop-gradient selects alignment directions that induce implicit contrastive separation, making self-supervised learning more stable and efficient.
TAC 2017 · System report
The system turns unstructured text into a knowledge base by extracting entity relations and linking them to supporting evidence through neural models and targeted rules.
Contributed baseline model evaluation and benchmarking as part of the NAVER Cloud HyperCLOVA X team.
Oct 2025–May 2026
Residency Program Participant · World Best LLM Project
Benchmarked vision-language and language models, contributed HyperCLOVA X baseline evaluations, and investigated information flow through audio-language model components.
Dec 2021–Aug 2022
Applied Scientist Intern
Developed anomaly detection for semiconductor process time series to support root-cause analysis of manufacturing defects.
Aug 2020–Feb 2021
Machine Learning Researcher
Developed deep learning models for financial time-series forecasting and downstream decision signals.
Ph.D. student in Artificial Intelligence
Kim Jaechul Graduate School of AI · Advisor: Prof. Jaesik Choi
2018–2020
M.S. in Computer Science and Engineering
Advisor: Prof. Jaesik Choi
2014–2018
B.S. in Electrical and Computer Engineering
Invited talk · AI EXPO KOREA, KAIST AI Technology Briefing
KAIST XAI Tutorial Series
KAIST XAI Tutorial Series · RISE and SISE
KAIRI Internship Program · KAIST
AI702 · KAIST
Samsung DS
UNIST
Reviewer: NeurIPS 2026; ICML 2026 Workshop on Mechanistic Interpretability; BMVC 2026; ICCV 2025; CVPR 2025.