🏔️ Jaehoon Hahm

Hi! I’m a 3rd year Physics Ph.D. Candidate at the University of Illinois, Urbana-Champaign, advised by Bryan K. Clark. I am currently collaborating with Ge Liu and Romit R. Choudhury.
Previously, I completed my M.S. at the Graduate School of Data Science, Seoul National University, advised by Joonseok Lee.
Before that, I completed my B.S. studying Physics and Mathematics at Seoul National University.

I’m interested in ML + Quantum, particularly using techniques in Geometric Deep Learning, and Generative Modeling.
I like to work in broad areas in the intersection of Physics, Differential Geometry, and ML.
I'm currently working on:

Optimal transport for local geometry preserving Diffusion/Flow/Energy-based models
Discrete Flow Matching for materials/molecules/proteins.
Foundational Neural Quantum States, Subspace Diagonalization

I am excited to work as an Research Scientist Intern at Meta Reality Labs this year (2026)! I am currently working on Generative models and Self-supervised Learning with Jogendra N. Kundu.

Feel free to contact me via email for any collaboration! Email: jh141@illinois.edu

Preprints

TILT: Test-Time Reward Alignment via Distribution Tilting for Compositional Generation

Debottam Dutta, Jaehoon Hahm , Jianchong Chen, Romit Roy Choudhury

Preprint, 2026

We present TILT, a training-free framework for compositional text-to-image generation via test-time reward alignment. We interpret compositional failures as overlap modes between joint and single-concept distributions, and define a pure-mode reward that favors samples where all concepts are jointly present while remaining close to the pretrained model.

Optimal Crystal Flow for Fast and Reliable Crystal Generation

Seunghoon Yi, Youngwoo Cho, Jaehoon Hahm , Jeongwoo Shin, Soo Kyung Kim, Jaegul Choo, Joonseok Lee, Hongkee Yoon

Preprint, 2026

De novo crystal generation aims to propose favorable candidates over the vast joint combinatorial space of atomic species and coordinates, accelerating materials discovery. In this work, we introduce Optimla Crystal Flow (OCFlow), a flow matching framework for crystal generation that constructs locally optimal probability paths in coordinate space through per-structure coupling tailored to crystalline systems.

ShadowFM: Geometric Approaches for Learning Quantum Many-body States with Flow Matching on Classical Shadows

Jaehoon Hahm, Bryan Clark

Preprint, 2026

We introduce Shadow Flow Matching, a novel generative framework for quantum state learning that combines classical shadow representations with Riemannian flow matching.

Paper

Publications

Equivariant Latent Alignment via Flow Matching under Group Symmetries

Sunghyun Kim*, Jaehoon Hahm*, Jeongwoo Shin, Joonseok Lee

ICML (43rd International Conference on Machine Learning) 2026

Geometry-aware generative models and novel view synthesis approaches have shown strong potential to improve visual fidelity and consistency. Our Residual Latent Flow is a flow matching-based framework that corrects the misaligned latents to improve equivariance relation upon specifc group actions.

Paper

SummDiff: Generative Modeling of Video Summarization with Diffusion

Kwanseok Kim*, Jaehoon Hahm*, Sumin Kim, Jinhwan Sul, Byung-Hak Kim, Joonseok Lee

ICCV (20th IEEE/CVF International Conference on Computer Vision) 2025, Highlight (263/11,239 = 2.3%)

We approach a video summarization problem using generative modeling framework, motivated from the diverse and subjective characteristic innate to the summarization task.

Paper

Generative Modeling of Quantum Distribution with Functional Flow Matching

Jaehoon Hahm*, Tak Hur*, Joonseok Lee, Daniel K. Park

QTML (8th International Conference on Quantum Techniques in Machine Learning) 2024

We use Functional Flow Matching to learn the distribution of quantum states via Wigner function mapping.

Paper

Isometric Representation Learning for Disentangled Latent Space of Diffusion Models

Jaehoon Hahm*, Junho Lee*, Sunghyun Kim, Joonseok Lee

ICML (41st International Conference on Machine Learning) 2024

We introduce a method to leverage isometric representation learning, motivated from Riemannian geometry, to obtain a smoother and disentangled latent space for diffusion mdoels.

Paper Code