Jaemoo Choi

Postdoc @ Georgia Tech

Profile

Hi! I’m a postdoctoral researcher at Georgia Tech, working at FLAIR Lab, with Prof. Yongxin Chen. I am very fortunate to be part of his group and to work with a supportive and inspiring community of researchers. I also have the pleasure of collaborating actively with Prof. Joonseok Lee and his research group. Prior to that, I received my Ph.D. in Mathematical Sciences and B.S. in Mathematics Education from Seoul National University.

My research focuses on the fundamental algorithms for generative AI (GenAI) and their applications across vision, language, and scientific domains (e.g., molecular generation). I am particularly interested in (discrete) diffusion models, flow-based methods, and large language models (LLMs) and its applications, as well as their connections to control and dynamical systems. Now, I am actively seeking full-time opportunities.

Contact: jchoi843 [at] gatech [dot] edu
Follow: Google Scholar | jaemoo-choi

Updates
[04/2026] 7 papers are accepted to ICML.
  • Rethinking RL4DM: Efficient fine-tuning of diffusion models through ELBO-based policy optimization
  • QUATRO: Reward-based fine-tuning of LLMs with exact trust-region-bounded policy optimization
  • DMPO (Spotlight 2.2%): Fine-tuning diffusion LLMs through cross-entropy policy optimization combined with ELBO estimation
  • DASBS: A discrete version of Adjoint Matching and ASBS. A first-order oracle for discrete diffusion sampler
  • GSBoG: Generalized Schr\"odinger Bridge Sampler on graphs
  • ASBM: Scalable diffusion-based generative modeling through ASBS
  • MetaDNS: Meta-dynamics Discrete Neural Sampler (preprint coming soon)
[01/2026] PDNS is accepted to ICLR.
  • PDNS: Proximal Diffusion Neural Sampler, a proximal cross-entropy-based diffusion sampler for both continuous and discrete state spaces
[09/2025] Three papers, ASBS (Oral 0.3%), NAAS, MDNS, accepted to NeurIPS.
  • ASBS: Adjoint Schrödinger Bridge Sampler, an efficient reward-based diffusion-based sampler framework. Generalization of Adjoint Sampler
  • NAAS: Non-equilibrium Annealed Adjoint Sampler. General Schr\"odinger Bridge Sampler with Annealed Dynamics
  • MDNS: Mask Diffusion Neural Sampler, masked diffusion neural sampler via stochastic optimal control
[05/2025] Two papers, OTP, UOT-UPC, accepted to ICML.
[01/2025] Two papers, U-NOTB, DIOTM, accepted to ICLR.
[09/2024] I have joined FLAIR, Georgia Tech as a Postdoc.
[05/2024] Scalable Wasserstein Gradient Flow for Generative Modeling through Unbalanced Optimal Transport is accepted to ICML.
[01/2024] Analyzing and Improving Optimal-Transport-based Adversarial Networks is accepted to ICLR.
[09/2023] Unbalanced Optimal Transport Model (UOTM) is accepted to NeurIPS.
Selected Publications
(∗ Equal contribution, * Core contributors, † Equal advising. See Google Scholar for the full list.)
  1. Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design
    Jaemoo Choi∗, Yuchen Zhu∗, Wei Guo, Petr Molodyk, Bo Yuan, Jinbin Bai, Yi Xin, Molei Tao, Yongxin Chen
    International Conference on Machine Learning (ICML), 2026
    arxiv
  2. QUATRO: Query-Adaptive Trust Region Policy Optimization for LLM Fine-tuning
    Doyeon Lee, Eunyi Lyou, Hyunsoo Cho†, Sookyung Kim†, Joonseok Lee†, Jaemoo Choi†
    International Conference on Machine Learning (ICML), 2026
    arxiv
  3. Discrete Adjoint Schrödinger Bridge Sampler
    Wei Guo∗, Yuchen Zhu, Xiaochen Du, Juno Nam, Yongxin Chen†, Rafael GĂłmez-Bombarelli†, Guan-Horng Liu†, Molei Tao†, Jaemoo Choi∗
    International Conference on Machine Learning (ICML), 2026
    arxiv
  4. Generalized Schrödinger Bridge on Graphs
    Panagiotis Theodoropoulos, Juno Nam, Evangelos Theodorou†, Jaemoo Choi†
    International Conference on Machine Learning (ICML), 2026
    arxiv
  5. Efficient Generative Modeling beyond Memoryless Diffusion via Adjoint Schrödinger Bridge Matching
    Jeongwoo Shin, Jinhwan Sul, Jaewoong Choi†, Joonseok Lee†, Jaemoo Choi†
    International Conference on Machine Learning (ICML), 2026
    arxiv
  6. Proximal Diffusion Neural Sampler
    Wei Guo∗, Jaemoo Choi∗, Yuchen Zhu, Molei Tao, Yongxin Chen
    International Conference on Learning Representations (ICLR), 2026
    arxiv
  7. Adjoint Schrödinger Bridge Sampler
    Guan-Horng Liu*, Jaemoo Choi*, Yongxin Chen, Benjamin K. Miller, Ricky T. Q. Chen*
    Advances in Neural Information Processing Systems (NeurIPS), 2025   [Oral 0.3%]
    arxiv
  8. Non-equilibrium Annealed Adjoint Sampler
    Jaemoo Choi*, Yongxin Chen, Molei Tao, Guan-Horng Liu*
    Advances in Neural Information Processing Systems (NeurIPS), 2025
    arxiv
  9. Overcoming Spurious Solutions in Semi-Dual Neural Optimal Transport: A Smoothing Approach for Learning the Optimal Transport Plan
    Jaemoo Choi, Jaewoong Choi†, Dohyun Kwon†
    International Conference on Machine Learning (ICML), 2025
    arxiv
  10. Robust barycenter estimation using semi-unbalanced neural optimal transport
    Milena Gazdieva∗, Jaemoo Choi∗, Alexander Kolesov, Jaewoong Choi, Petr Mokrov, Alexander Korotin
    International Conference on Learning Representations (ICLR), 2025
    arxiv | code
  11. Scalable Wasserstein Gradient Flow for Generative Modeling through Unbalanced Optimal Transport
    Jaemoo Choi∗, Jaewoong Choi∗, Myungjoo Kang
    International Conference on Machine Learning (ICML), 2024
    arxiv | code
  12. Generative Modeling through the Semi-dual Formulation of Unbalanced Optimal Transport
    Jaemoo Choi∗, Jaewoong Choi∗, Myungjoo Kang
    Advances in Neural Information Processing Systems (NeurIPS), 2023
    arxiv | code