Seungjun Lee

Research Staff, Brookhaven National Laboratory

I build machine learning models for complex physical systems. Three threads run through my work: neural operator surrogates that replace expensive PDE solvers, diffusion-based methods for inverse problems on experimental data, and uncertainty quantification for scientific prediction.

At Brookhaven I work on surrogate models for relativistic hydrodynamics, phase retrieval from X-ray diffraction, and test-time scaling for robot policies. I train at scale on NERSC Perlmutter in PyTorch and JAX, and review for NeurIPS, ICML, ICLR, AAAI, and TMLR.

A model that explains a family of related systems is worth more than one that fits a single narrow dataset. That is how governing equations work in physics, and it is the property I try to build into learning-based models.

Publications

Efficient Neural Field Learning via Adaptive Coverage and Focused Sampling Guang Zhao, Xihaier Luo, Huan-Hsin Tseng, Seungjun Lee, Shinjae Yoo, Yihui Ren, Wei Xu NeurIPS, 2026 paper
Efficient Surrogate Modeling for Fast Hydrodynamic Evolution of the Quark-Gluon Plasma Seungjun Lee, Yeonju Go, Bjoern Schenke, Chun Shen, Yihui Ren, Jin Huang, Maria Chamizo-Llatas ICML Workshop on AI for Physics, 2026 paper
PDEfuncta: Spectrally-Aware Neural Representation for PDE Solution Modeling Minju Jo, Woojin Cho, Uvini Balasuriya Mudiyanselage, Seungjun Lee, Noseong Park, Kookjin Lee NeurIPS, 2025 paper
SCENT: Robust Spatiotemporal Learning for Continuous Scientific Data via Scalable Conditional Neural Fields David Keetae Park, Xihaier Luo, Guang Zhao, Seungjun Lee, Miruna Oprescu, Shinjae Yoo ICML, 2025 paper
Generalizable Implicit Neural Representations via Parameterized Latent Dynamics for Baroclinic Ocean Forecasting Guang Zhao, Xihaier Luo, Seungjun Lee, Yihui Ren, Shinjae Yoo, Luke Van Roekel, Balu Nadiga, Sri Hari Krishna Narayanan, Yixuan Sun, Wei Xu ICLR Workshop on Tackling Climate Change with ML, 2025 paper
Large-Scale Training in Neural Compact Models for Accurate and Adaptable MOSFET Simulation Chanwoo Park, Seungjun Lee, Junghwan Park, Kyungjin Rim, Jihun Park, Seonggook Cho, Jongwook Jeon, Hyunbo Cho IEEE Journal of the Electron Devices Society, 2024 paper
Inducing Point Operator Transformer: A Flexible and Scalable Architecture for Solving PDEs Seungjun Lee, Taeil Oh AAAI, 2024 paper code
Mesh-Independent Operator Learning for Partial Differential Equations Seungjun Lee ICML Workshop on AI for Science, 2022 paper
Identifying Physical Law of Hamiltonian Systems via Meta-Learning Seungjun Lee, Haesang Yang, Woojae Seong ICLR, 2021 NeurIPS Workshop on Machine Learning and the Physical Sciences, 2020 paper code workshop
Zero-Shot Single-Microphone Sound Classification and Localization in a Building via the Synthesis of Unseen Features Seungjun Lee, Haesang Yang, Hwiyong Choi, Woojae Seong IEEE Transactions on Multimedia, 2021 paper data
Type/Position Classification of Inter-Floor Noise in Residential Buildings with a Single Microphone via Supervised Learning Hwiyong Choi, Haesang Yang, Seungjun Lee, Woojae Seong EUSIPCO, 2020 paper
Classification of Inter-Floor Noise Type/Position via Convolutional Neural Network-Based Supervised Learning Hwiyong Choi, Haesang Yang, Seungjun Lee, Woojae Seong Applied Sciences, 2019 paper

Selected work

Neural operator surrogates for quark-gluon plasma hydrodynamics. Replacing 3+1D viscous relativistic hydrodynamic solvers with a tensorized Fourier neural operator, trained on large simulation ensembles for event-by-event heavy-ion physics at RHIC.
Diffusion-based phase retrieval for X-ray diffraction. Plug-and-play diffusion priors for inverting ptychography and Bragg coherent diffraction data, evaluated on real experimental measurements rather than simulation alone.
Test-time scaling for vision-language-action policies. Studying when additional inference-time compute improves flow-matching robot policies, and what limits it. Embodied AI LDRD at Brookhaven.
Neural compact models for semiconductor devices. Large-scale training of MOSFET surrogates for SPICE-compatible circuit simulation at Alsemy.

Background

2025 –: Research Staff, Computing and Data Sciences, Brookhaven National Laboratory
2024 – 2025: Research Associate, Brookhaven National Laboratory
2022 – 2024: AI Researcher, Alsemy, Seoul
2017 – 2022: PhD, Seoul National University. Advisor: Woojae Seong
2012 – 2017: BS, Seoul National University

Awards

Korean Mathematical Olympiad, Gold (middle school, 2008), Silver (middle school, 2006) and Bronze (high school, 2010)
Korean Physics Olympiad, Silver (middle school, 2007)
Outstanding Graduate Student Award, BK21 Program, Ministry of Education (2022)

Reviewing

NeurIPS 2025, 2026. ICML 2026. ICLR 2025, 2026. AAAI 2026, 2027. TMLR 2025.