Taeseong Yoon

Ph.D. Candidate in Industrial & Systems Engineering, KAIST

bigstar0423@kaist.ac.kr  |  Google Scholar  |  GitHub  |  LinkedIn
Engineering Data Science Lab, KAIST, Daejeon, South Korea

Print / save as PDF

Research summary

My research primarily focuses on uncertainty quantification (UQ) for deep learning. I develop scalable, theoretically grounded frameworks to ensure reliable and trustworthy predictions, with a core emphasis on evidential deep learning, efficient Bayesian inference, out-of-distribution detection, and model robustness under distribution shifts. My current work focuses on post-hoc UQ for foundation models and, more broadly, on UQ for agentic AI systems.

Beyond foundational UQ, my broader research interests encompass statistical machine learning for complex structured and industrial data, where I advance methodologies for multivariate time series anomaly detection, learning under incomplete and imbalanced data, causal inference, and tabular foundation models.

Education

KAIST (Korea Advanced Institute of Science and Technology)
Ph.D. Candidate, Industrial & Systems Engineering
Sep 2022 – Feb 2028 (expected)
Daejeon, South Korea
  • Advisor: Heeyoung Kim
  • Research focus: Uncertainty quantification for deep learning
KAIST (Korea Advanced Institute of Science and Technology)
B.S. in Industrial & Systems Engineering
Mar 2018 – Aug 2022
Daejeon, South Korea

Publications

* denotes equal contribution; bold denotes the candidate.

Refereed publications

  1. GT-PSSM: Unified Probabilistic Framework for Stochastic Dynamics Modeling and Dependency Learning in Multivariate Time Series Anomaly Detection
    Wonmo Koo, Jaeyeong Lee, Taeseong Yoon, Heeyoung Kim
    ACM International Conference on Information and Knowledge Management (CIKM 2026)
    arXiv
  2. Courtroom Analogy: New Perspective on Uncertainty-Aware Classification
    Taeseong Yoon, Heeyoung Kim
    International Conference on Machine Learning (ICML 2026), PMLR 306
    PMLRarXivCode
  3. Knowledge-Assisted Multi-Graph Dependency Learning for Multivariate Time Series Anomaly Detection in Multi-Stage Industrial Processes
    Jaeyeong Lee*, Taeseong Yoon*, Wonmo Koo, Heeyoung Kim
    IEEE Transactions on Automation Science and Engineering (IEEE TASE), 2026
    IEEE XplorearXiv
  4. Uncertainty Estimation by Flexible Evidential Deep Learning
    Taeseong Yoon, Heeyoung Kim
    Advances in Neural Information Processing Systems (NeurIPS 2025)
    ProceedingsarXivCode
  5. Uncertainty Estimation by Density-Aware Evidential Deep Learning
    Taeseong Yoon, Heeyoung Kim
    International Conference on Machine Learning (ICML 2024), PMLR 235
    PMLRarXivCode

Manuscripts under review

  1. Separating Prediction from Epistemic Uncertainty in Evidential Deep Learning
    Taeseong Yoon, Seungjun Choi, Heeyoung Kim
    International Conference on Learning Representations (ICLR 2027), under review
    OpenReview
  2. Strong Views for Representation, Weak Views for Classification: Rethinking Augmentation in Class-Imbalanced Semi-Supervised Learning
    Taemin Park, Taeseong Yoon, Heeyoung Kim
    International Conference on Learning Representations (ICLR 2027), under review
  3. Causal Role-Aware Variable Selection for Continuous Treatment Effect Estimation
    Hyunsoo Cho*, Taeseong Yoon*, Insoo Jeong*, Heeyoung Kim
    International Conference on Artificial Intelligence and Statistics (AISTATS 2027), under review

Research projects

Uncertainty Quantification for Class-Imbalanced and Noisy Label Datasets
Project Lead (Ph.D. Researcher); PI: Prof. Heeyoung Kim
2026 – 2027
POSCO
  • Industry-initiated project proposed and funded by POSCO following the publication of DAEDL (ICML 2024), extending single-pass uncertainty quantification methods to class-imbalanced and noisy label settings in real-world industrial data.
  • Led the overall research direction and technical development, delivering uncertainty-aware learning frameworks that leverage predictive uncertainty for reliable model training under distribution shifts.
  • Keywords: evidential deep learning, Bayesian deep learning, single-pass uncertainty quantification, class imbalance, label noise, distribution shift.
Development of Causality-Based Interpretable Integrated Framework for Manufacturing Process Control
Graduate Student Researcher; PI: Prof. Heeyoung Kim
2023 – 2028
National Research Foundation of Korea (NRF)
  • Keywords: causal inference, time-series anomaly detection, manufacturing process automation.
Human-Centered Carbon-Neutral Global Supply Chain Research Center
Graduate Student Researcher; PI: Prof. Hyerim Bae
2023 – 2028
NRF, Engineering Research Center (ERC)
  • Keywords: time-series analysis, spatiotemporal forecasting, AIS (Automatic Identification System) data, maritime logistics.

Research & industry experience

Samsung Advanced Institute of Technology (SAIT)
Research Intern, Machine Learning Lab, AI Research Center
Jul 2023 – Aug 2023
Suwon, South Korea
  • Developed uncertainty-aware anomaly detection methods for semiconductor quality inspection.
Industrial Statistics Lab, KAIST
Undergraduate Research Intern
Jul 2021 – Aug 2021
Daejeon, South Korea
  • Studied Bayesian nonparametric modeling, with a focus on Dirichlet process Gaussian mixture models.
SK Hynix
Intern, QA Team, Mobile DRAM Engineering
Jan 2021 – Feb 2021
Icheon, South Korea
  • Analyzed bank-address patterns in mobile DRAM quality data to support fault detection and manufacturing defect analysis.

Honors & awards

Selected talks & presentations

Academic service

Teaching & leadership

Skills