When can we trust AI systems?  

We study AI systems under uncertainty and changing conditions, with a focus on understanding when their predictions and decisions can be trusted. 

As AI becomes part of increasingly important decisions, knowing when it can be relied on is as important as improving what it can do. 

Research

Uncertainty & Decision-Making

How can uncertainty, confidence, calibration, prediction risk, and abstention help us determine whether an AI prediction or decision can be trusted?

AI under Change

How do AI predictions and uncertainty change when data, models, tasks, or environments change?

Failure, Anomaly & Safety

How can we recognize unexpected or unsafe behavior, including out-of-distribution inputs, anomalies, hallucinations, and unreliable or manipulated data?

Application Areas

Foundation Models & LLMs · Physical AI & Robotics · Financial AI & Risk · Healthcare & BioAI · AI Security & Data Integrity

These domains provide diverse real-world settings in which uncertainty, distribution shift, and failure become critical to the reliable use of AI.

Current Research

Training, Representation, and OOD Detection

How do training choices change the representation geometry used by uncertainty and OOD detectors?

Uncertainty and Abstention in LLMs

What does a language model know about whether a question can be answered, and why does that knowledge not always lead to abstention?

Data Integrity and Poisoning Detection

How can suspicious training data be identified when clean reference data or known attack detectors are unavailable?

Recent News 

• [2026.09] Jinhyeok’s first paper was accepted to AACL-IJCNLP 2026 Main Conference, in his first semester as an M.S. student. Congrats Jinhyeok! 🎉

• [2026.01] “Do geopolitical shocks move the primary market? Evidence from U.S. Treasury auction yields”, co-authored with Jae-Gyu Song, was published in Applied Economics Letters (SSCI). Jae-Gyu, who worked on this research as an undergraduate student in Economics, is now pursuing graduate studies at Sogang University. Congratulations! 🎉 

Join Us 

We welcome students who want to grow as researchers in machine learning and trustworthy AI.

Students in our lab learn to formulate research questions, develop hypotheses, design rigorous experiments, interpret results, and communicate original research. We encourage students who aim to develop independent research for leading international AI/ML/NLP venues.

Research Training
Foundations → First Research Project → Top-Conference Research Training → Student-Led Research → Research Leadership

Target venues

ICLR · ICML · NeurIPS · AAAI · AISTATS · UAI · KDD

ACL · EMNLP · NAACL · EACL · AACL-IJCNLP · COLING

Track A. Representation Learning, OOD & Distribution Shift

Track B. Foundation Models: Reliability, Reasoning & Alignment

Track C. Data Integrity, Robustness & AI Security

Track D. Physical AI: Uncertainty, Decision-Making & Safety

                                                                                                                                old