Hyunjin Seo

Hyunjin Seo

I'm a Ph.D. student at KAIST, advised by Prof. Sungsoo Ahn. My research focuses on practical applications of geometric deep learning. In particular, I explore the integration of large language models (LLMs) with graph neural networks (GNNs) to enhance graph reasoning capabilities. Central to my work is enhancing graph data by decomposing inherent semantic relationships of binary connections between textual entities using LLMs. Additionally, I have developed pruning and calibration techniques tailored for GNNs to address the computational challenges and network under-confidence.

My work also extends into the domain of molecules, where I develop graph transformer for ground-state molecular conformation prediction through energy-based graph rewiring. Recently, I am designing a property-guided molecule diffusion model for controlled polymer generation.

NEWS


Jan. 2025   Paper accepted at ICLR '25
Dec. 2024   Paper accepted at AAAI '25
Oct. 2024   ReBind released on arXiv
May. 2024   RoSE released on arXiv
May. 2024   ICLR '24 @ Vienna, Austria
Jan. 2024   Paper accepted at ICLR '24
Jul. 2023   Paper accepted at ICCV '23
Mar. 2023   Started my M.S. & Ph.D. journey at KAIST

PUBLICATIONS


ReBind: Enhancing Ground-state Molecular Conformation Prediction via Force-Based Graph Rewiring

Taewon Kim*, Hyunjin Seo*, Sungsoo Ahn, Eunho Yang (*: equal contribution)

ICLR 2025   PDF

Towards Precise Prediction Uncertainty in GNNs: Refining GNNs with Topology-grouping Strategy

Hyunjin Seo, Kyusung Seo*, Joonhyung Park*, Eunho Yang (*: equal contribution)

AAAI 2025   PDF

Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large Language Models

Hyunjin Seo*, Taewon Kim*, June Yong Yang, Eunho Yang (*: equal contribution)

Preprint   PDF

TEDDY: Trimming Edges with Degree-based Discrimination strategY

Hyunjin Seo*, Jihun Yun*, Eunho Yang (*: equal contribution)

ICLR 2024   PDF

PC-Adapter: Topology-Aware Adapter for Efficient Domain Adaption on Point Clouds with Rectified Pseudo Label

Joonhyung Park, Hyunjin Seo, Eunho Yang

ICCV 2023   PDF

Efficient Subword Segmentation for Korean Language Classification

Hyunjin Seo*, Jeongjae Nam*, Minseok Kim* (*: equal contribution)

HCLT 2022

Analysis and Utilization of Hypergraph Neural Networks

Hyunjin Seo, Seongjun Yun, Jaewoo Kang

KCC 2021 (Recipient of the Best Undergraduate Award)   PDF

Relation-aware Pseudo-labeling for Link Prediction with Graph Neural Networks

Hyunjin Seo*, Seongjun Yun*, Buru Chang, Jaewoo Kang (*: equal contribution)

Under Review   PDF

EDUCATION


KAIST Logo

Korea Advanced Institute of Science and Technology (KAIST)

2023.02 - Present

  • M.S. & Ph.D. integrated student
  • Kim Jaechul Graduate School of Artificial Intelligence
KU Logo

Korea University

2018.02 - 2022.08

  • Bachelor’s degree
  • History, Artificial Intelligence (Double Major)

EXPERIENCE


AI Research Intern at Polymerize

2024.10 - Present

  • Headed an independent research team focused on multi-objective molecule generation, tackling a wide range of structures like polymers and organic molecules. Beyond my research, I offered machine learning consultation to internal domain experts, enhancing their projects with technical insights.

Teaching Assistant

2023.09 - 2023.12, 2024.03 - 2024.06

  • Machine Learning for Artificial Intelligence (KAIST)

Research Intern at Machine Learning and Intelligence Lab (MLILAB), KAIST

2022.04 - 2023.02

  • Tackled efficient domain adaptation for point clouds, critical due to distributional shifts in real-world applications, by developing GNN-based topology-aware adapters to seamlessly adapt models from source data to target local characteristics.

Research Intern at Data Mining and Information Systems Lab (DMIS), Korea University

2021.01 - 2022.02

  • Developed a pseudo-labeling approach for graph edge prediction task, assigning labels to disconnected node pairs using a hierarchical scoring mechanism.

PROJECTS


Structural Guidance for Enhancing Molecular Understanding in Large Language Models

2025.02 - Present

Samsung Electronics

Molecular Property Prediction via Denoising-based Graph Neural Networks

2024.06 - 2024.12

Samsung Electronics

Predicting Food Product Satisfaction using Large Language Models

2023.05 - 2023.11

LAB-EAT

AWARDS


Best Undergraduate Award

2021

Korea Computer Congress (KCC)

Top 2.7%

Great Honors Award

2019, 2020, 2021

Korea University

SERVICES


Conference Reviewer

  • International Conference on Learning Representations (ICLR)