Kyeongseon (Ella) Kim김경선

I am a Ph.D. student in the School of Computing at KAIST. I work with Prof. Tae-Hyun Oh in the AMI Lab.

My research explores how visual structure, interaction, and cultural context contribute to multimodal understanding. I uncover informative signals, such as semantic structure in vector graphics and implicit cues in user behavior, and design representations that help models and agents understand meaning and users’ needs. My goal is to make unfamiliar digital environments more approachable, helping people of all ages, abilities, and backgrounds explore and adapt with confidence.

Before my Ph.D., I was a research scientist at POSTECH, a research engineer at LG Electronics, and an AI education manager at Upstage, the company behind the Solar LLM. I founded and organize AGI Korea (formerly TensorFlow Korea), an AI community of 66,000+ members.

EMNLP 2026 Main

IGG: A Benchmark for Interactive GUI Grounding under Visibility Constraints

K. S. Kim*, J. Son*, T.-H. Oh

Conference on Empirical Methods in Natural Language Processing

Also presented at ICML 2026 Workshop on Compositional Learning: Safety, Interpretability, and Agents

Project

CVPR 2026

SVHalluc: Benchmarking Speech–Vision Hallucination in Audio-Visual Large Language Models

C. Zhang, K. S. Kim, C. Liu, T.-H. Oh

IEEE/CVF Conference on Computer Vision and Pattern Recognition

arXiv Project

WACV 2026

mEOL: Training-Free Instruction-Guided Multimodal Embedder for Vector Graphics and Image Retrieval

K. S. Kim, S.-E. Baek, J.-M. Lee, T.-H. Oh

IEEE/CVF Winter Conference on Applications of Computer Vision

Round 1 paper (6.4% acceptance rate)

Best Student Presentation Award, IPIU 2025

Also presented at ICCV 2025 Workshop on Graphic Design Understanding and Generation (GDUG)

arXiv Project Poster

ICML 2026 Workshop

CuPS: Measuring Cultural Preference Signatures in LLM/VLM Agents and Their Steering by Profile Memories

K. S. Kim, G. Kim, J. Chang, H. Kim, T.-H. Oh

Workshop on Culture × AI: Evaluating AI as a Cultural Technology

AAAI 2025

Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior

H. Lee, K. S. Kim, B.-K. Kwon, T.-H. Oh

AAAI Conference on Artificial Intelligence

Qualcomm Innovation Fellowship 2025 winning paper

Paper arXiv Project

Applied Sciences 2019

Personalized Online Live Video Streaming Using Softmax-Based Multinomial Classification

K. S. Kim, D. Kwon, J. Kim, A. Mohaisen

Applied Sciences, Vol. 9, No. 11, Article 2297

Paper

ICAIIC 2019

Hardness on Style Transfer Deep Learning for Rococo Painting Masterpieces

K. S. Kim, D. Kim, J. Kim

International Conference on Artificial Intelligence in Information and Communication

ICUFN 2018

Detecting Selfish Backoff Attack in IEEE 802.15.4 CSMA/CA Using Logistic Classification

J. Kim, K. S. Kim

International Conference on Ubiquitous and Future Networks

IEEE SECON 2018

Recipient-oriented Transaction for Preventing Double Spending Attacks in Private Blockchain

H. Lee, M. Shin, K. S. Kim, Y. Kang, J. Kim

IEEE International Conference on Sensing, Communication, and Networking

IEEE ICASSP 2018

Self-Adaptive Machine Learning Operating Systems for Security Applications

K. S. Kim, D. Kwon, Y. Kim, J. Kim, J. Kim

IEEE International Conference on Acoustics, Speech and Signal Processing