Prof.Mingyu Kim김민규 교수김민규Mingyu Kim

Contact
Phone 02-910-4753
E-mail mgyukim@kookmin.ac.kr
E-mail mgyukim@kookmin.ac.kr
Office사무실생활관D동 8층 8호 (02-910-4753)
Education학력
Ph.D. KAIST, Artificial Intelligence (2024)
M.S. KAIST, Ocean System Engineering (2014)
B.S. Konkuk University, Industrial Engineering (2012)
M.S. KAIST, Ocean System Engineering (2014)
B.S. Konkuk University, Industrial Engineering (2012)
Career경력
2026-Present Assistant Professor, Dept. of AI, Kookmin University
2024-2026 Postdoctoral Fellow, UBC Computer Science
2025-2026 CIFAR AI Safety Postdoctoral Fellow
2014-2019 Research Engineer, Samsung Heavy Industries
2024-2026 Postdoctoral Fellow, UBC Computer Science
2025-2026 CIFAR AI Safety Postdoctoral Fellow
2014-2019 Research Engineer, Samsung Heavy Industries
2026-현재 국민대학교 AI학과 조교수
2024-2026 UBC 컴퓨터과학과 박사후연구원
2025-2026 CIFAR AI 안전성 박사후연구원
2014-2019 삼성중공업 연구원
2024-2026 UBC 컴퓨터과학과 박사후연구원
2025-2026 CIFAR AI 안전성 박사후연구원
2014-2019 삼성중공업 연구원
Research Overview연구 개요
My research focuses on probabilistic generative modeling, including diffusion models, flow matching, and neural processes. I am particularly interested in developing safe and controllable generative AI through negative guidance and inference-time steering. I also study reliable adaptation of foundation models using Bayesian post-training methods to improve generalization, as well as multimodal and 3D intelligence through vision-language learning and neural representations.
확산 모델, 플로 매칭, 신경 과정 등을 포함한 확률적 생성 모델링을 중심으로 합니다. 특히 네거티브 가이던스와 추론 시점 조향 기법을 활용하여 안전하고 제어 가능한 생성형 인공지능을 개발하는 데 관심이 있습니다. 또한 베이지안 사후 학습을 통해 파운데이션 모델을 안정적으로 적응시키고 일반화 성능을 향상하는 방법을 연구하고 있습니다. 이와 함께 비전 언어 학습과 신경 표현을 기반으로 멀티모달 및 3차원 지능을 구현하는 연구도 수행하고 있습니다.
Research Areas연구 분야
- Probabilistic Generative Modeling
- Safe and Controllable Generative AI
- Reliable Foundation Model Adaptation
- Multimodal and 3D Intelligence
- 확률적 생성 모델링
- 안전하고 제어 가능한 생성형 인공지능
- 신뢰성 높은 파운데이션 모델 적응
- 멀티모달 및 3차원 지능
Major Achievements주요 연구 성과
- Safety-Guided Flow (SGF) for safe generation - ICLR 2026 Oral (Top 1.2%)
- Training-free safe denoisers for diffusion models (NeurIPS 2025)
- Neural Processes with Stochastic Attention (ICLR 2022)
- Safety-Guided Flow (SGF) for safe generation - ICLR 2026 Oral (Top 1.2%)
- Training-free safe denoisers for diffusion models (NeurIPS 2025)
- Neural Processes with Stochastic Attention (ICLR 2022)
Honors & Awards수상
- Area Chair, ICLR2027
- Top Area Chair, NeurIPS 2026
- Top Reviewer, NeurIPS 2025
- Best Reviewer, AISTATS 2025
- 2nd Winner, ML4CO Challenge, NeurIPS 2021
Recent Publications주요 논문
- Minjoo Choi, Jaekyeong Lee, Stein Ove Erikstad, Mingyu Kim, "LLM-as-UI: A preliminary exploration of fine-tuned language models as intelligent interfaces in modular ship design," Ocean Engineering, Vol. 356, pp. 125344, 2026 원문 ↗Paper ↗
- M. Kim, Y.-H. Kim, and M. Park, "Safety-Guided Flow (SGF): A Unified Framework for Negative Guidance in Safe Generation," ICLR (Oral), 2026 원문 ↗Paper ↗
- M. Kim, D. Kim, A. Yusuf, S. Ermon, and M. Park, "Training-Free Safe Denoisers For Safe Use of Diffusion Models," NeurIPS, 2025
- M. Kim, J. Ko, and M. Park, "Bayesian Principles Improve Prompt Learning In Vision-Language Models," AISTATS, 2025
- J.-S. Kim, M. Kim, G.-U. Kim, T.-H. Oh, and J.-H. Kim, "Factorized Multi-Resolution HashGrid for Efficient Neural Radiance Fields," IEEE RA-L / ICRA, 2024
- M. Kim, J.-S. Kim, S.-Y. Yun, and J.-H. Kim, "Synergistic Integration of Coordinate Network and Tensorial Feature for Improving Neural Radiance Fields from Sparse Inputs," ICML, 2024
- M. Kim, J. Lee, L. Choi, and M. Choi, "PolarGAN: Creating Realistic Arctic Sea Ice Concentration Images with User-Defined Geometric Preferences," Engineering Applications of Artificial Intelligence, 2023
- M. Kim, K. R. Go, and S.-Y. Yun, "Neural Processes with Stochastic Attention: Paying more attention to the context dataset," ICLR, 2022
