Diffusion Fine-Tuning via Reparameterized Policy Gradient of the Soft Q-Function
Kang, H.*, Lee, J.*, Shin, W.*, Om, K., & Park, J. (2026). "Diffusion Fine-Tuning via Reparameterized Policy Gradient of the Soft Q-Function." ICLR.
CV
Personal Information
- Birth: 18th September 1999, Republic of Korea
- Nationality: Korean
- Languages: Korean (native), English (fluent)
- Github: https://github.com/umkiyoung
- Linkedin: https://www.linkedin.com/in/kiyoung-om
- Contact: se99an@gmail.com, se99an@kaist.ac.kr
- Phone: +82-10-2575-5334
Research Interests
- Traffic simulation for autonomous driving
- Closed-loop covariate shift in sequential decision-making systems
- Generative models for optimization and control
- Alignment of generative models via search and fine-tuning
Education
- B.S. in Industrial Management Engineering, Korea University, Mar. 2018 - Feb. 2024
- GPA: 4.02/4.5 (3.57/4.0), Major GPA: 4.19/4.5 (3.75/4.0)
- M.S. in Data Science, Korea Advanced Institute of Science and Technology (KAIST), Mar. 2024 - Feb. 2026
- Graduate School of Data Science
- Advisor: Prof. Jinkyoo Park, System Intelligence Lab
Research Experience
- Research Intern, NAVER LABS, Mar. 2026 - Aug. 2026
- Working on closed-loop covariate shift in traffic simulation for autonomous driving
- Studying robust simulation and learning methods for autonomous vehicle policies interacting with dynamic traffic environments
- AI Research Assistant, Samsung Electronics-KAIST AI Industry-Academia Collaboration, Mar. 2025 - Feb. 2026
- Developed AI-enhanced supply chain management optimization using subgradient lambda prediction
- Improved optimization speed and efficiency in complex SCM problems through machine learning approaches
Awards and Honors
- Academic Excellence Award, Korea University (2022)
- Capstone Excellence Award, Korea University (2022)
- Topic: Understanding the Lifecycle and Gap Technologies of Lithium-Ion Batteries
- Commander’s Commendation (2020)
- Speech Contest on Practicing the Core Values of the Army and Showcasing Examples
Extracurricular Activities
- KUBIG, Regular Member (July 2022 - July 2023)
- Zero-shot object detection using CLIP with Grad-CAM application
- Vehicle hazard detection using YOLOv7
- Prediction of breast cancer lymph node metastasis AI contest
- Algorithm trading/Backtesting with FinRL
- Solar Cell Lab - AI Industry-Academia Collaboration Program, AI application advisor (April 2023 - July 2023)
- Prediction of Weighted-Average reflection of solar cells using panel images
- Optimal Experimental Date and Efficiency Prediction of Perovskite Solar Cells based on Weather Forecast
Publications
Diffusion Alignment as Variational Expectation-Maximization
Lee, J., Kim, M., Choi, S., Song, I., Yun, S., Kang, H., Shin, W., Yun, T., Om, K., & Park, J. (2026). "Diffusion Alignment as Variational Expectation-Maximization." ICLR.
Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization (CiBO)
Om, K.*, Sim, K.*, Yun, T.*, Kang, H., & Park, J. (2025). "Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization." NeurIPS Workshop on Structured Probabilistic Inference & Generative Modeling (Oral).
Posterior Inference with Diffusion Models for High-dimensional Black-box Optimization (DiBO)
Yun, T.*, Om, K.*, Lee, J., Yun, S., & Park, J. (2025). "Posterior Inference with Diffusion Models for High-dimensional Black-box Optimization." ICML; FPI @ ICLR Workshop.