Minjae Kwon

Charlottesville, VA

I am a PhD candidate in the Department of Computer Science at the University of Virginia, advised by Prof. Lu Feng. Previously, I received a PhD in mathematics from Kyungpook National University.

My research asks one question: how can learning systems adapt to changing environments while maintaining safety at runtime? Training-time safety guarantees can break when deployment conditions differ from training. My work addresses this problem in three directions:

Adaptive runtime safety — safety wrappers that infer changes in the dynamics at deployment and adjust their safety margin under uncertainty, using basis-adaptive neural ODEs and conformal prediction without retraining the policy.
Safe in-context adaptation — frozen policies that adapt their strategy from interaction history while respecting a safety budget, using methods such as safe algorithm distillation and latent Q-barrier filtering.
Compositional safe adaptation — agents that execute unseen compositions of temporally structured tasks (specified in linear temporal logic) under distribution shift.

I study these problems in safety-critical domains, including robotic control in MuJoCo simulation environments and healthcare applications such as personalized diabetes management using physiological simulators.

news

Sep 2026	Paper accepted to NeurIPS 2026: Latent Q-Barrier Shielding for Safe In-Context Reinforcement Learning.
Jul 2026	Invited talk at the Korea Institute for Advanced Study (KIAS): “Safe Decision-Making When the World Changes”.
May 2026	Two papers accepted to ICML 2026: Safe In-Context Reinforcement Learning and Safety Generalization Under Distribution Shift in Safe RL: A Diabetes Testbed.
May 2026	Received the ICML 2026 Gold Reviewer Award.
Jul 2025	Received a UAI 2025 conference scholarship.

selected publications

* Equal contribution.

ICML

Safe In-Context Reinforcement Learning

Amir Moeini^*, Minjae Kwon^*, Alper Kamil Bozkurt, Yuichi Motai, Rohan Chandra, Lu Feng, and Shangtong Zhang

International Conference on Machine Learning (ICML), 2026

Bib PDF

@inproceedings{moeini2026safeincontext,
  title = {Safe In-Context Reinforcement Learning},
  author = {Moeini, Amir and Kwon, Minjae and Bozkurt, Alper Kamil and Motai, Yuichi and Chandra, Rohan and Feng, Lu and Zhang, Shangtong},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2026},
}

ICML

Safety Generalization Under Distribution Shift in Safe Reinforcement Learning: A Diabetes Testbed

Minjae Kwon, Josephine Lamp, and Lu Feng

International Conference on Machine Learning (ICML), 2026

Bib PDF

@inproceedings{kwon2026safetygeneralization,
  title = {Safety Generalization Under Distribution Shift in Safe Reinforcement Learning: A Diabetes Testbed},
  author = {Kwon, Minjae and Lamp, Josephine and Feng, Lu},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2026},
}

NeurIPS

Latent Q-Barrier Shielding for Safe In-Context Reinforcement Learning

Minjae Kwon^*, Amir Moeini^*, Shangtong Zhang, and Lu Feng

Conference on Neural Information Processing Systems (NeurIPS), 2026

Bib PDF

@inproceedings{kwon2026safeicrllatentbarrier,
  title = {Latent Q-Barrier Shielding for Safe In-Context Reinforcement Learning},
  author = {Kwon, Minjae and Moeini, Amir and Zhang, Shangtong and Feng, Lu},
  booktitle = {Conference on Neural Information Processing Systems (NeurIPS)},
  year = {2026},
}

Preprint

Adaptive Shielding for Safe Reinforcement Learning under Hidden-Parameter Dynamics Shifts

Minjae Kwon, Tyler Ingebrand, Ufuk Topcu, and Lu Feng

Preprint, 2026

arXiv Bib PDF

@article{kwon2026adaptiveshielding,
  title = {Adaptive Shielding for Safe Reinforcement Learning under Hidden-Parameter Dynamics Shifts},
  author = {Kwon, Minjae and Ingebrand, Tyler and Topcu, Ufuk and Feng, Lu},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {cs.LG},
}

UAI

Adaptive Reward Design for Reinforcement Learning

Minjae Kwon, Ingy ElSayed-Aly, and Lu Feng

Conference on Uncertainty in Artificial Intelligence (UAI), 2025

Reward design for linear temporal logic specifications

Bib PDF

@inproceedings{kwon2025adaptiverewarddesign,
  title = {Adaptive Reward Design for Reinforcement Learning},
  author = {Kwon, Minjae and ElSayed-Aly, Ingy and Feng, Lu},
  booktitle = {Conference on Uncertainty in Artificial Intelligence (UAI)},
  year = {2025},
  note = {Reward design for linear temporal logic specifications},
}