우리 연구팀은 자연어처리(NLP)와 멀티모달 AI 분야에서 활발한 연구를 수행하며, AAAI 2026, EMNLP 2025, ACL 2025, NAACL 2025 등 주요 국제 학회에 지속적으로 연구 성과를 발표하고 있습니다. 또한 한국연구재단 우수신진연구·중견연구 과제 선정 등 다양한 외부 연구비 프로젝트를 수행하며 안정적인 연구 환경을 구축하고 있습니다.
다음 두 전공 트랙에서 함께 연구할 학부 연구생과 대학원생(석·박사)을 모집합니다.
일반대학원 AI학과: LLM 기반 언어 이해·추론, 편향·안전성 분석, 데이터 증강, 윤리·정확성 평가, 다국어 모델링 등 다양한 NLP 연구를 수행합니다. 최근에는 LLM 에이전트 기반 평가·검증, 언어적 편향 완화, 구조적 검색·추론 모델링 등 최신 연구도 적극적으로 진행하고 있습니다.
첨단영상대학원 영상학과: 텍스트–이미지 융합 모델, 약지도 기반 이미지 해석, 비전–언어 표현 학습, 도메인 일반화, 멀티모달 밸런스 조정 등 영상·언어 결합 기반의 핵심 문제를 다룹니다. ECCV, WACV 등에서 꾸준히 연구 성과를 내고 있으며 의료·과학 데이터 기반의 응용 멀티모달 연구도 병행하고 있습니다.
NLP 또는 멀티모달 AI 연구에 관심 있는 분은 간단한 자기소개와 이력서를 ybkim85@cau.ac.kr(김영빈 교수)에게 보내주세요.
(2026) We are delighted to announce that two of our papers have been accepted to AAAI 2026. Congratulations to all authors on this achievement!
"Easy to Learn, yet Hard to Forget: Towards Robust Unlearning under Bias" (Main Technical Track)
"RefLens: End-to-End Evidence-Grounded Citation Verification with LLM Agents" (Demonstration Program)
(2026) Our paper has been accepted to the AAAI 2026 Workshop on AI for Research (AI4Research):
"Position on LLM-Assisted Peer Review: Addressing Reviewer Gap through Mentoring and Feedback"
(2025) We are pleased to announce that our research project, 대화형 인공지능을 활용한 존엄한 임종에 관한 대화 촉진 에이전트, conducted in collaboration with Seoul National University Hospital, has been selected for funding under the Excellent Young Researcher Grant (우수신진연구) program by the National Research Foundation of Korea (한국연구재단, NRF).
(2025) We are delighted to announce that four of our papers have been accepted to EMNLP 2025. Congratulations to all authors on this excellent achievement!
"CoBA: Counterbias Text Augmentation for Mitigating Various Spurious Correlations via Semantic Triples" (Main Conference)
"LLM Agents at the Roundtable: A Multi-Perspective and Dialectical Reasoning Framework for Essay Scoring" (Findings of EMNLP 2025, conducted in collaboration with NC AI)
"Beyond Single-User Dialogue: Assessing Multi-User Dialogue State Tracking Capabilities of Large Language Models" (Findings of EMNLP 2025)
"From Ground Trust to Truth: Disparities in Offensive Language Judgments on Contemporary Korean Political Discourse" (Findings of EMNLP 2025)
(2025) Our paper, "Query, Decompose, Compress: Structured Query Expansion for Efficient Multi-Hop Retrieval," has been accepted to CIKM 2025.
(2025) Our papers have been accepted to the following workshops:
ICCV 2025 Workshop on Curated Data for Efficient Learning (CDEL): "VolDoGer: LLM-assisted Datasets for Domain Generalization in Vision-Language Tasks"
COLM 2025 First Workshop on Optimal Reliance and Accountability in Interactions with Generative Language Models (ORIGen): "Medal Matters: Probing LLMs’ Failure Cases Through Olympic Rankings"
MICCAI 2025 Workshop on Applications of Medical AI (AMAI): "XTag-CLIP: Robust and Reliable Thyroid Scar Analysis with Limited Data via Cross-Attention"
(2025) We are delighted to announce that two of our papers have been accepted to the ACL 2025 Main Conference. Congratulations to all authors for this excellent achievement! The accepted papers are:
"Delving into Multilingual Ethical Bias: The MSQAD with Statistical Hypothesis Tests for Large Language Models"
"Plug-in and Fine-tuning: Bridging the Gap between Small Language Models and Large Language Models"
(2025) We are pleased to share that our paper, "Making Sense of Korean Sentences: A Comprehensive Evaluation of LLMs through KoSEnd Dataset," has been accepted to the ACL 2025 Student Research Workshop (SRW).
(2025) Prof. YoungBin Kim has been awarded the Best Reviewer Award at AISTATS 2025. (link)
(2025) We are pleased to announce that our research project, Neural Journalism++: 논리·윤리·공정성 강화 언어 모델 기반 지능형 저널리즘 연구, has been selected for funding under the Consolidator Grants (중견연구) program by the National Research Foundation of Korea (한국연구재단, NRF).
(2025) We are delighted to announce that our paper, led by Junehyoung Kwon, has been accepted to NAACL 2025 Main Conference as an ORAL PRESENTATION: "See-Saw Modality Balance: See Gradient, and Sew Impaired Vision-Language Balance to Mitigate Dominant Modality Bias"
(2025) Prof. YoungBin Kim was honored for excellence in the Fall 2024 Course Evaluation.
(2024) Prof. YoungBin Kim will serve as an Area Chair for ACL Rolling Review (ARR).
(2025) We are pleased to share that our paper has been accepted to the AAAI 2025 (Demo Track): "SummPilot: Bridging Efficiency and Customization for Interactive Summarization System" (News)
(2024) We are delighted to announce that three of our papers have been accepted to the EMNLP 2024 Main Conference. Congratulations to all the authors on this outstanding achievement! The accepted papers are (News):
(2024) We are delighted to announce that our paper, led by Soojin Jang, has been accepted to ECCV 2024. Congratulations to Soojin on this outstanding achievement! - "DIAL: Dense Image-text ALignment for Weakly Supervised Semantic Segmentation" (Link)
(2024) We are delighted to announce that our paper, led by Seunguk Yu, has been accepted to the Findings of NAACL 2024. Congratulations, Seunguk, on this excellent work! - "Don't be a Fool: Pooling Strategies in Offensive Language Detection from User-Intended Adversarial Attacks" (Link)
(2024) Our paper has been accepted to COLING 2024. - "Enhancing Effectiveness and Robustness in a Low-Resource Regime via Decision-Boundary-aware Data Augmentation" (Link)
(2024) We are pleased to inform that our paper, led by Juhwan Choi, has been accepted to the Findings of EACL 2024. Congratulations to Juhwan on this achievement! - "GPTs Are Multilingual Annotators for Sequence Generation Tasks" (Link)
(2024) We'd like to share the acceptance of a paper by Juhwan Choi to the EACL 2024 Student Research Workshop. - "AutoAugment Is What You Need: Enhancing Rule-based Augmentation Methods in Low-resource Regimes" (Link)
(2024) We are happy to share that our paper, led by JuneHyoung Kwon, has been accepted to WACV 2024. Well done, JuneHyoung! - "Learning to Detour: Shortcut Mitigating Augmentation for Weakly Supervised Semantic Segmentation" (Link)
(2023) We are pleased to announce that our paper, led by Jungmin Yun, has been accepted to the Findings of EMNLP 2023 and will be presented in-person at the SpLU-RoboNLP workshop. Congratulations to Jungmin on this achievement! - "Focus on the Core: Efficient Attention via Pruned Token Compression for Document Classification" (Link)
Our lab is dedicated to solving challenges and exploring opportunities within low-resource environments. By focusing on efficient and innovative approaches in data augmentation, natural language processing (NLP), and multimodal AI, we aim to make AI technologies more accessible and impactful. Our mission is to develop solutions that are practical, inclusive, and capable of addressing real-world constraints, particularly in scenarios with limited resources.
- We are advancing the frontiers of multilingual and low-resource language processing by developing inclusive technologies that enhance accessibility and usability across diverse languages and dialects. Our work includes reducing linguistic bias, improving grammatical accuracy, and enabling more equitable language technologies.
- To overcome the challenges of data scarcity, we design innovative data augmentation techniques for both text and images. These strategies include methods that progressively improve model robustness and performance, ensuring adaptability to low-resource environments while minimizing resource consumption.
- Our research focuses on optimizing the efficiency of AI systems through robust training methods and model compression techniques. By addressing computational and data limitations, we aim to ensure that AI systems remain sustainable and practical in real-world applications, even with constrained resources.
Through these focus areas, our lab contributes to making AI technologies more accessible and equitable, fostering solutions that address the global challenges of inclusivity and resource scarcity. By combining innovation with practical impact, we aim to drive the development of AI that benefits communities worldwide and meets the pressing needs of diverse, real-world applications.