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 β€” specifically, the linguistic robustness of VLA models and multimodal policy learning. More recently, I have also been exploring efficient VLA from a perception perspective.

profile photo

News
2026.04 πŸ”₯ LIBERO-Para featured as #3 on Hugging Face Daily Papers! (26.04.07)
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
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 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] πŸ”₯ #3 on HF Daily Papers (26.04.07)

Domestic
VLA Pre-Planning Vision-Language-Action λͺ¨λΈμ˜ Pre-Planning LLM λͺ¨λΈμ— λ”°λ₯Έ μ„±λŠ₯ 및 평가 효율
Chanyoung Kim, Dahuin Jung
KCC 2025 (Korea Computer Congress), KIISE, Jul 2025
[paper]
RL LLM Prompt κ°•ν™”ν•™μŠ΅μ— κΈ°λ°˜ν•œ LLM ν”„λ‘¬ν”„νŠΈ μ‘°μ • 기법에 λŒ€ν•œ μ΅œμ‹  연ꡬ 동ν–₯ 뢄석
Chanyoung Kim, Haneul Pyeon, Byungchan Ko, Dahuin Jung
KICS Journal (정보와톡신), Vol.42(4), Mar 2025
[paper]
RL LLM Prompt Conf κ°•ν™”ν•™μŠ΅μ— κΈ°λ°˜ν•œ LLM ν”„λ‘¬ν”„νŠΈ μ‘°μ • 기법에 λŒ€ν•œ μ΅œμ‹  연ꡬ 동ν–₯ 뢄석
Chanyoung Kim, Haneul Pyeon, Byungchan Ko, Dahuin Jung
KICS Fall Conference, Nov 2024
[paper]
LLM Quantization λ ˆμ΄μ–΄ 별 μ–‘μžν™” 였λ₯˜ μ΅œμ ν™” 기법 쀑심 κ±°λŒ€ μ–Έμ–΄ λͺ¨λΈμ— λŒ€ν•œ ν›ˆλ ¨ ν›„ μ–‘μžν™” μ΅œμ‹  기술 뢄석
Byungchan Ko, Hyunoh Kang, Chanyoung Kim, Dahuin Jung
KICS Fall Conference, Nov 2024
[paper]
Self-supervised Tabular λΉ„ μˆœμ°¨ν˜• μ •ν˜• 데이터에 λŒ€ν•œ μžκ°€ 지도 ν•™μŠ΅ 기법에 κ΄€ν•œ μ΅œμ‹  기술 동ν–₯ 뢄석
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 Γ— NDE+X Lab (Hanyang Univ.) Β· 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 Γ— NDE+X Lab (Hanyang Univ.) Β· 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 Β· Collaborative research

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

Template adapted from Jon Barron.