Jeongjae Lee

I am a Master’s student at KAIST AI, advised by Prof. Jong Chul Ye. I received my B.S. in Electrical and Computer Engineering from Seoul National University (Summa Cum Laude).

Research Focus

I develop algorithms at the intersection of Reinforcement Learning and Diffusion/Flow Matching, specifically focusing on RL-based fine-tuning of generative models. I am also interested in generative models for RL policies, and inference-time steering of generative models.

Selected Work

• [ICLR 2026] PCPO: Proportionate Credit Policy Optimization — I proposed a novel policy gradient objective that stabilizes the fine-tuning of diffusion and flow-matching models, solving the high-variance credit assignment problem in preference alignment.

Background

Prior to my focus on generative control, I worked on optimization in high-stakes environments. At DeepMetrics, I engineered predictive features for Optimal Decision Trees, to automate ICU ventilator control. I also co-authored Foldseek (Nature Biotechnology), where I optimized LDDT alignment score computations using spatial hashing and SIMD instructions, resulting in a 13x speedup over the baseline.

Outside the lab, I am a competitive dancer (specializing in Popping) and gym enthusiast.