Sunghyun Baek

About Me

I am currently working at Samsung Research as a Research Engineer. I received my Ph.D. in Electrical Engineering from KAIST, where I was advised by Prof. Junmo Kim. During my Ph.D., I worked on efficient AI, including network pruning, token pruning, and test-time adaptation, and on LLM safety, including unlearning and red-teaming. These days I am mainly interested in LLM agents, and more broadly in building AI systems that are efficient, reliable, and practical to deploy.

Research Interests

Machine Learning
Computer Vision
Natural Language Processing

News

Sep. 2026One paper accepted to NeurIPS 2026 (Negative Result Track, Spotlight).New
Sep. 2026One paper accepted to ACCV 2026.New
Aug. 2026Three papers accepted to EMNLP 2026 (two Main, one Findings).
May. 2026One paper accepted to ICML 2026 (Spotlight).
Mar. 2026One paper accepted to ACL 2026 (Main).
Jan. 2026One paper accepted to ICLR 2026.
2025One paper accepted to Pattern Analysis and Applications.
2024One paper accepted to Knowledge-Based Systems.
2022One paper accepted to ICIP 2022.
2021One paper accepted to AAAI 2021.

Publications

Imitation Dominates Reinforcement: Direct In-Context RL Is Closer to ICL Than RL

Minchan Kwon, Seunghee Koh, Sunghyun Baek, Minsung Bae, Junmo Kim

NeurIPS, 2026 (Negative Result Track, Spotlight)
VGGT-Bridge: Beyond Sequential Pose Graphs via Coarse-Stride Skip Edges

Sungjae Choi, Hanna Bae, Sunghyun Baek, Junmo Kim

ACCV, 2026
IWP: Token Pruning as Implicit Weight Pruning in Large Vision Language Models

Dong-Jae Lee^*, Sunghyun Baek^* and Junmo Kim (^*equally contributed)

EMNLP, 2026 (Findings)

arXiv
<\think> Doesn't Stop Reasoning: Analysis of Spurious CoT Termination

Seunghee Koh, Sungjae Choi, Minchan Kwon, Sunghyun Baek, Junmo Kim

EMNLP, 2026 (Main)

arXiv
Direct Diversity Optimization for Diverse Successful Trajectories in Preference Post-Training

Junwon Ko, Dongjae Lee, Minchan Kwon, Sunghyun Baek, Junmo Kim

EMNLP, 2026 (Main)

arXiv
Stable-GFlowNet: Toward Diverse and Robust LLM Red-Teaming via Contrastive Trajectory Balance

Minchan Kwon, Sunghyun Baek, Minseo Kim, Jaemyung Yu, Dongyoon Han, Junmo Kim

ICML, 2026 (Spotlight, top 2.2%)

arXiv ICML page
Forget What Matters, Keep the Rest: Selective Unlearning of Informative Tokens

Seunghee Koh, Sunghyun Baek, Youngdong Kim, Junmo Kim

ACL, 2026 (Main)

arXiv ACL Anthology
IMSE: Intrinsic Mixture of Spectral Experts Fine-tuning for Test-Time Adaptation

Sunghyun Baek, Jaemyung Yu, Seunghee Koh, Minsu Kim, Hyeonseong Jeon, Junmo Kim

ICLR, 2026

arXiv OpenReview
Random Pruning
at Initialization

Prior knowledge of layer-specific pruning numbers guarantees effective random pruning at initialization

Minju Jung, Sunghyun Baek, Yunho Jeon, Junmo Kim

Pattern Analysis and Applications, 2025

Paper
AI-KD: Adversarial learning and Implicit regularization for self-Knowledge Distillation

Hyungmin Kim, Sungho Suh, Sunghyun Baek, Daehwan Kim, Daun Jeong, Hansang Cho, Junmo Kim

Knowledge-Based Systems, 2024

arXiv Paper
Pruning
Metrics

Which Metrics for Network Pruning: Final Accuracy? Or Accuracy Drop?

Donggyu Joo, Sunghyun Baek, Junmo Kim

ICIP, 2022
Linearly Replaceable Filters for Deep Network Channel Pruning

Donggyu Joo, Eojindl Yi, Sunghyun Baek, Junmo Kim

AAAI, 2021

Paper Code

Preprints

Info3R: Information Adaptive Test-Time Update in 3D Reconstruction

Sunghyun Baek, Hanna Bae, Minchan Kwon, Junmo Kim

Under review
Supervised Dimension Contrastive Learning

Jaemyung Yu, Sunghyun Baek, Jongsuk Kim, Junmo Kim

Under review
DAM: Domain-Aware Module for Multi-Domain Dataset Condensation

Jaehyun Choi, Gyojin Han, Dongjae Lee, Sunghyun Baek, Junmo Kim

Under review

arXiv

Education

Mar. 2020 – Aug. 2026 Ph.D. in Electrical Engineering, KAIST
Advisor: Prof. Junmo Kim
Mar. 2018 – Feb. 2020 M.S. in Electrical Engineering, KAIST
Advisors: Prof. Junmo Kim and Prof. Jinwoo Shin
Mar. 2014 – Feb. 2018 B.S. in Electrical Engineering, Sogang University