cv
Research Interests
My research combines AI, physics, and neuroscience to uncover how large language models learn, store, and revise knowledge. Using representation geometry and causal interventions, I study memory, model editing, and machine unlearning, seeking simple mechanisms behind complex behavior and testable hypotheses inspired by biological memory. I aim to extend this approach to continual reinforcement learning by identifying how rewards preserve existing capabilities and selectively consolidate experience, with the goal of building safer, more reliable, and more efficient AI systems.
Education
- 2016 - 2022
Doctor of Philosophy (PhD) in Physics
Korea Advanced Institute of Science and Technology (KAIST), Daejeon, South Korea
- Supervised by Prof. Hawoong Jeong.
- Dissertation: "Nonequilibrium Statistical Physics Study using Deep Learning." [pdf]
- 2011 - 2015
Bachelor of Science (BS) in Physics
Seoul National University (SNU), Seoul, South Korea
- Major in Physics and Minor in Computer Science & Engineering.
Employment History
- 2024 - Current
Postdoctoral Researcher
Max Planck Institute for Security and Privacy (MPI-SP)
- 2022 - 2024
Postdoctoral Researcher
Institute for Basic Science (IBS)
- Data Science Group, Center for Mathematical and Computational Sciences.
- Hosted by Prof. Meeyoung Cha (Chief Investigator).
- Sep. 2017 - Dec. 2017
Machine Learning Intern
Samsung Electronics
- Collaborated with Daniel Kim (Senior Data Scientist).
- Improved anomaly image classification tasks via distributed multi-GPU training methods of Keras and Spark.
- Implemented a distributed image searching framework to detect similar patterns in images through Elasticsearch.