Chanyoung Kim

I am an undergraduate researcher at the Human-Centered AI Lab, Soongsil University, where I work with Prof. Dahuin Jung. Since February 2026, I have also been an undergraduate researcher at the Human-Centered AI Lab, Chung-Ang University, under the same advisor.

I started my research journey in the summer of 2024, shortly after entering university, and have since been involved in diverse projects spanning vision-language-action models and natural language processing.

My research interests center on Vision-Language-Action (VLA) models and NLP — specifically, the linguistic robustness of VLA models and multimodal policy learning. More recently, I have also been exploring efficient VLA from a perception perspective.

News

2026.03	First-author preprint LIBERO-Para: A Diagnostic Benchmark and Metrics for Paraphrase Robustness in VLA Models released on arXiv.
2026.02	Joined the Human-Centered AI Lab at Chung-Ang University as an undergraduate researcher.
2025.07	Paper on VLA pre-planning published at KCC 2025 (KIISE).
2025.03	Survey paper on RL-based LLM prompt tuning published in KICS Journal.
2024.11	Presented three papers at KICS Fall Conference 2024.
2024.06	Joined the Human-Centered AI Lab at Soongsil University as an undergraduate researcher.
2024.03	Started B.S. in Computer Science & Engineering at Soongsil University.

Research Experience

Undergraduate Researcher
Feb 2026 – Present

Human-Centered AI Lab, Chung-Ang University
Advised by Prof. Dahuin Jung

Research on efficient VLA, NLP, and perception
First-author paper: LIBERO-Para: A Diagnostic Benchmark and Metrics for Paraphrase Robustness in VLA Models (arXiv 2026)
Co-authored a paper on LLM continual unlearning, submitted to CoLM 2026 (second author)

Undergraduate Researcher
Jun 2024 – Feb 2026

Human-Centered AI Lab, Soongsil University
Advised by Prof. Dahuin Jung

Research on VLA models and NLP, including linguistic robustness and multimodal policy learning
Co-authored a paper on VLM pruning, submitted to ECCV 2026 (second author)
Co-authored a paper on ID-preserving multi-view generation, submitted to SIGGRAPH 2026 (second author)
Published 5 domestic papers (1 journal, 4 conference)

International

LIBERO-Para: A Diagnostic Benchmark and Metrics for Paraphrase Robustness in VLA Models
Chanyoung Kim, Minwoo Kim, Minseok Kang, Hyunwoo Kim, Dahuin Jung
arXiv preprint, 2026
[project page] [arxiv]

Domestic

	Vision-Language-Action 모델의 Pre-Planning LLM 모델에 따른 성능 및 평가 효율 Chanyoung Kim, Dahuin Jung KCC 2025 (Korea Computer Congress), KIISE, Jul 2025 [paper]
	강화학습에 기반한 LLM 프롬프트 조정 기법에 대한 최신 연구 동향 분석 Chanyoung Kim, Haneul Pyeon, Byungchan Ko, Dahuin Jung KICS Journal (정보와통신), Vol.42(4), Mar 2025 [paper]
	강화학습에 기반한 LLM 프롬프트 조정 기법에 대한 최신 연구 동향 분석 Chanyoung Kim, Haneul Pyeon, Byungchan Ko, Dahuin Jung KICS Fall Conference, Nov 2024 [paper]
	레이어 별 양자화 오류 최적화 기법 중심 거대 언어 모델에 대한 훈련 후 양자화 최신 기술 분석 Byungchan Ko, Hyunoh Kang, Chanyoung Kim, Dahuin Jung KICS Fall Conference, Nov 2024 [paper]
	비 순차형 정형 데이터에 대한 자가 지도 학습 기법에 관한 최신 기술 동향 분석 Haneul Pyeon, Chanyoung Kim, Jian Kim, Dahuin Jung KICS Fall Conference, Nov 2024 [paper]

Research Projects

Dec 2024 – Present	Time-Series AI for Semiconductor Quality Inspection Human-Centered AI Lab × NDE+X Lab (Hanyang Univ.) · Sole student researcher under two faculty advisors Developing a model-agnostic, low-data training solution for challenging defect cases in GPU core components
Sep 2025 – Present	Efficient On-Device Semiconductor Inspection Human-Centered AI Lab × NDE+X Lab (Hanyang Univ.) · Collaborative research Exploring low-cost, efficient methods for non-destructive semiconductor testing on edge devices
Jan 2026 – Present	Medical LLM Agent Human-Centered AI Lab · Collaborative research

Awards

2024	Gold Prize, Computer System Basic Design (Capstone) Course, Soongsil University