publications

2026

1. 

   Distribution-based Regularization for Reward Fine-tuning Diffusion and Flow Models

   Romeo Passaro^*, Jeongjae Lee^*, Jong Chul Ye, and 4 more authors

   arXiv preprint, 2026

   arXiv Code

2. 

   GeoFix: Geometry Fixing for 4D Consistent Generation

   Yeobin Hong, Hyojun Go, Dohun Lee, Jeongjae Lee, and 6 more authors

   arXiv preprint, 2026

   arXiv

3. 

   Development and Validation of VentPilot: An AI-based Recommendation System for Mechanical Ventilation

   Hong Yeul Lee^*, Gaon An^*, Yeonwoo Jeong, Edward J. Schenck, Ho Geol Ryu, Brian W. Pickering, Vitaly Herasevich, Daniel A. Diedrich, Seungyong Moon, Hyun-Lim Yang, Hyung-Chul Lee, Wun Hur, Jaewon Choi, Seokjun Lee, Jeongjae Lee, and 7 more authors

   Journal of Intensive Care, 2026

   DOI

4. 

   Aligning Few-Step Generative Models by Amortizing Sample-based Variational Inference

   Jaewoo Lee^*, Hyeongyu Kang^*, Dohyun Kim, Kyuil Sim, Woocheol Shin, Minsu Kim, Taeyoung Yun, Jeongjae Lee, and 4 more authors

   In Advances in Neural Information Processing Systems (NeurIPS), 2026

   Also at the ICML 2026 Workshop on Structured Probabilistic Inference and Generative Modeling.

   arXiv Code

5. 

   Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models

   Jeongjae Lee^*, Jinho Chang^*, Jeongsol Kim^†, and 1 more author

   In Advances in Neural Information Processing Systems (NeurIPS), 2026

   Also at the ICML 2026 Workshop on Structured Probabilistic Inference and Generative Modeling.

   Best Paper arXiv Code Website

   Best Paper Award at the ICML 2026 SPIGM Workshop (one of 2 recipients among 270 submissions).

6. 

   PCPO: Proportionate Credit Policy Optimization for Aligning Image Generation Models

   Jeongjae Lee, and Jong Chul Ye

   In International Conference on Learning Representations (ICLR), 2026

   HTML Code

2024

1. 

   Fast and accurate protein structure search with Foldseek

   Michel van Kempen^*, Stephanie S. Kim^*, Charlotte Tumescheit, Milot Mirdita, Jeongjae Lee, and 3 more authors

   Nature Biotechnology, 2024

   Abs DOI Code

   As structure prediction methods are generating millions of publicly available protein structures, searching these databases is becoming a bottleneck. Foldseek aligns the structure of a query protein against a database by describing tertiary amino acid interactions within proteins as sequences over a structural alphabet. Foldseek decreases computation times by four to five orders of magnitude with 86%, 88% and 133% of the sensitivities of Dali, TM-align and CE, respectively.