Integrated BS-MS Course
SoHyeon Kim
sohyeon967@gmail.com
Research Interests
•
Efficient AI
•
Vision-Language Model
Publication
(* indicates first author, † indicates corresponding author)
International Conference Papers
[C4] Focus Matters: Attention-Value Dynamics for Hallucination Mitigation in Vision-Language Models
Sohyeon Kim*, Sang Yeon Yoon, Kyeongbo Kong†
NeurIPS, 2026.
[C3] Uncertainty-Guided Graph Formulation via MWIS for Token Pruning in LVLMs
Jouwon Song*, Sohyeon Kim*, Kyeongbo Kong†
CVPR Findings, 2026.
[C2] AgilePruner: An Empirical Study of Attention and Diversity for Adaptive Visual Token Pruning in Large Vision-Language Models
Changwoo Baek*, Jouwon Song*, Sohyeon Kim*, Kyeongbo Kong†
ICLR 2026, 2026.
Initial version was presented NeurIPSW, 2025
[C1] Efficient Token Pruning for Large Vision-Language Models in Multi-turn Scenarios
Changwoo Baek*, Sohyeon Kim*, Jouwon Song*, Kyeongbo Kong†
IEEE Asia Pacific Conference on Circuits and Systems (APCCAS), 2025.
International Workshop Papers
[W2] Mitigating Visual Hallucinations for Reliable Multimodal Agents
Sohyeon Kim*, Sang Yeon Yoon, Kyeongbo Kong†
ICML Workshop (Agents in the Wild: Safety, Security, and Beyond), 2026.
[W1] Failure-Aware Query Refinement for Reliable Open-Vocabulary Home-Robot Perception
Daun Jeong*, Sohyeon Kim*, Jouwon Song, Kyeongbo Kong†
ICML Workshop (Agents in the Wild: Safety, Security, and Beyond), 2026.
Domestic Conference Papers
1.
김소현*, 백창우*, 송주원*, 공경보†, “이미지 복잡도와 환각 생성 관점에서 본 대규모 비전-언어 모델의 시각 토큰 프루닝 분석”, 대한전자공학회 추계학술대회, 2025년 11월. (장려상)
2.
김소현*, 윤상연, 공경보†, “대규모 비전-언어 모델의 객체 환각 완화를 위한 비전 인코더 Attention-Value 동역학 분석”, 대한전자공학회 하계학술대회, 2026년 06월. (우수학생논문상)
Domestic Journal Papers
1.
백창우*, 김소현*, 송주원*, 백광렬, 공경보†, “멀티모달 LLM에서의 비전 인코더 계층별 엔트로피 변화를 활용한 시각 토큰 프루닝, 방송공학회논문지, 2026년 1월.
Project
•
홈로봇 환경을 고려한 VLM 기반시각 인식 기술 개발 및 상황 인지·추론 성능 개선, LG Electronics, Korea /공동연구원