Junghyun Lee

📍 Seoul, Republic of Korea

📧 ejunghyun@ewha.ac.kr

🏫 Ewha Womans University

I am an integrated M.S.-Ph.D. student in AI at Ewha Womans University, advised by Prof. Junhyug Noh in the PAI Lab.

Research Interests

My research spans multimodal AI and human-centered computing, with a focus on:

Multimodal Emotion Recognition — blended emotion, audio-visual fusion
Sign Language Recognition — robustness under occlusion, AI fairness for deaf communities
Multimodal Reasoning, Active Learning, Domain Adaptation
Child Development & Developmental Psychology
Early Detection of MCI and Dementia

Background

I bring 15+ years of industry experience spanning software engineering, R&D leadership, and startup management. I served as CTO at BtoWIN Co., Ltd., leading AI-driven patent analytics and commercializing ipradar.kr. As Founder & CEO of GemPack Co., Ltd., I launched 9 AR/AI-based children’s learning applications and secured multiple government ICT grants. Prior to that, I worked as a server engineer at LINE PLAY (LINE Group) and Entiq Games, building large-scale backend systems serving millions of users.

I also hold a B.S. in Computer Science and B.A. in Multimedia Studies (Ewha Womans University), a B.A. in Child Studies, and an EMBA (Aalto University / aSSIST), and previously served as a KOICA volunteer computer teacher in Asunción, Paraguay, where I taught CS fundamentals at a national high school — an experience that shaped my interest in human-centered AI and educational technology.

컴퓨터공학 학사(복수전공: 멀티미디어학), 아동학 학사 및 EMBA 학위를 보유하고 있으며, KOICA 35기 해외봉사단으로 파라과이 아순시온에서 고등학교 컴퓨터 교사로 활동한 경험이 이후 인간 중심 AI 및 에듀테크 연구에 대한 관심으로 이어졌습니다.

Selected Publications

FGW
Ordering Matters: Rank-Aware Selective Fusion for Blended Emotion Recognition

Junghyun Lee, Hyunseo Kim, Hanna Jang, and 1 more author

In BlEmoRe Challenge at International Conference on Automatic Face and Gesture Recognition (FGW), 2026, 2nd Place Winner in Multimodal Blended Emotion Recognition

Abs arXiv Bib HTML PDF

Blended emotion recognition is challenging because emotions are often expressed as mixtures of subtle and overlapping multimodal cues rather than a single dominant signal. We propose a rank-aware multi-encoder framework that selectively combines complementary representations from diverse pre-extracted video and audio encoders.
@inproceedings{lee2026ordering, title = {Ordering Matters: Rank-Aware Selective Fusion for Blended Emotion Recognition}, author = {Lee, Junghyun and Kim, Hyunseo and Jang, Hanna and Noh, Junhyug}, booktitle = {BlEmoRe Challenge at International Conference on Automatic Face and Gesture Recognition (FGW), 2026}, location = {2nd Place Winner in Multimodal Blended Emotion Recognition}, url = {https://arxiv.org/abs/2605.21417}, }
ICLRW
Random Is Hard to Beat: Active Selection in Online DPO with Modern LLMs

Giyeong Oh, Junghyun Lee, Jaehyun Park, and 3 more authors

I Can’t Believe It’s Not Better (ICBINB) Workshop at International Conference on Learning Representations (ICLRW), 2026

Abs arXiv Bib PDF Code

Modern LLMs inherit strong priors from web-scale pretraining, which can limit the headroom of post-training data-selection strategies. While Active Preference Learning (APL) seeks to optimize query efficiency in online DPO, simple Random sampling proves a surprisingly formidable baseline.
@article{oh2026random, title = {Random Is Hard to Beat: Active Selection in Online {DPO} with Modern {LLMs}}, author = {Oh, Giyeong and Lee, Junghyun and Park, Jaehyun and Yu, Youngjae and Bae, Wonho and Noh, Junhyug}, journal = {I Can’t Believe It’s Not Better (ICBINB) Workshop at International Conference on Learning Representations (ICLRW), 2026}, url = {https://arxiv.org/abs/2604.02766}, }