Embedded Files
Our lab investigates how to design an intelligence system based on learning the underlying patterns or rules in large-scale data.
Our lab investigates how to design an intelligence system based on learning the underlying patterns or rules in large-scale data.
We study data mining and machine learning algorithms to learn useful knowledge from various types of data including but not limited to image, text, and time-series data. Also, we find a way to create data-driven intelligence systems for manufacturing, service, and healthcare industries.
We study data mining and machine learning algorithms to learn useful knowledge from various types of data including but not limited to image, text, and time-series data. Also, we find a way to create data-driven intelligence systems for manufacturing, service, and healthcare industries.
We are looking for passionate students who are interested in machine learning and deep learning. If you are interested in joining our lab, send an e-mail with your curriculum vitae, research statement, and transcript to shwang [at] seoultech.ac.kr.
We are looking for passionate students who are interested in machine learning and deep learning. If you are interested in joining our lab, send an e-mail with your curriculum vitae, research statement, and transcript to shwang [at] seoultech.ac.kr.
Recent Research Topics / Research Area
Recent Research Topics / Research Area
With the emergence of foundation models such as large language models (LLMs) and large multimodal models (LMMs) trained on vast amounts of data, our lab currently explores methods for adapting these models to specific applications and enhancing the reliability and safety of AI systems. Specifically, we conduct research on unlearning techniques for removing specific information from LLMs, safety alignment to guide LLMs toward producing safe responses, concept erasing to eliminate certain concepts from image generation models, and personalizatio techniques for customizing image generation models.
With the emergence of foundation models such as large language models (LLMs) and large multimodal models (LMMs) trained on vast amounts of data, our lab currently explores methods for adapting these models to specific applications and enhancing the reliability and safety of AI systems. Specifically, we conduct research on unlearning techniques for removing specific information from LLMs, safety alignment to guide LLMs toward producing safe responses, concept erasing to eliminate certain concepts from image generation models, and personalizatio techniques for customizing image generation models.
ML/DL Algorithms
ML/DL Algorithms
- AI safety for foundation models such as LLMs and VLMs
- Machine unlearning for generative models
- Confidence calibration and uncertainty quantification of deep neural networks
- Unknown unknowns detection: misclassification/out-of-distribution detection
- Domain generalization by incorporating structural priors
ML/DL Applications
ML/DL Applications
- Medical image analysis
- Manufacturing system
- Recommender system
Recent Publications
Recent Publications
[CVPR 2025] Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations
[CVPR 2025] Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations
Jeonghyeon Kim and Sangheum Hwang
We propose a multi-modal fine-tuning (MMFT) approach to improve out-of-distribution detection (OoDD) in vision-language models. By addressing the modality gap between image and text embeddings through cross-modal alignment, our method enhances the utilization of pretrained knowledge, achieving state-of-the-art OoDD performance on ImageNet-1k benchmarks.
[CVPR 2025] APT: Adaptive Personalized Training for Diffusion Models with Limited Data
[CVPR 2025] APT: Adaptive Personalized Training for Diffusion Models with Limited Data
Jungwoo Chae*, Jiyoon Kim*, Jaewoong Choi, Kyungyul Kim, and Sangheum Hwang (*equal contribution)
We introduce Adaptive Personalized Training (APT), a framework for personalizing diffusion models with limited data while mitigating overfitting and preserving prior knowledge. APT achieves this through adaptive training adjustment, representation stabilization, and attention alignment, ensuring semantic coherence and high-quality image generation.
Jihyo Kim*, Seulbi Lee*, and Sangheum Hwang (*equal contribution)
We assess the out-of-distribution detection (OoDD) capabilities of large vision-language models (LVLMs), addressing concerns about their trustworthiness and reliability. To enhance OoDD performance, we propose Reflexive Guidance (ReGuide), a self-guided prompting approach that utilizes image-adaptive concept suggestions. Our experiments show that ReGuide improves both image classification and OoDD performance in LVLMs.
Jihyo Kim*, Sungchul Kim*, Seungwon Seo, Bumsoo Kim, Daejeong Mun, Hoonjae Lee, and Sangheum Hwang (*equal contribution)
We propose an advanced contrastive learning method for facial skin representation learning, leveraging metadata-driven multi-instance contrastive relationships to enhance performance in skin disorder analysis. Our approach improves classification accuracy and transferability across different skin lesion datasets, demonstrating the potential of application-specific batch configurations in self-supervised learning.
NEWS
NEWS
(2025.03) [Welcome] Seungeun Lee joined our lab as an M.S. student.
(2025.03) [Welcome] Yeongbi Na joined our lab as an intern.
(2025.02) [IC] "Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations" with Jeonghyeon Kim was accepted to CVPR 2025 and will be presented in Nashville, USA in June 2025.
(2025.02) [IC] "APT: Adaptive Personalized Training for Diffusion Models with Limited Data" with Jungwoo Chae*, Jiyoon Kim*, Jaewoong Choi, and Kyungyul Kim (all from LG CNS) was accepted to CVPR 2025 and will be presented in Nashville, USA in June 2025. (*equal contribution)
(2025.02) [IJ] "Metadata Enriched Multi-Instance Contrastive Learning for High-Quality Facial Skin Visual Representations" with Jihyo Kim*, Sungchul Kim*, Seungwon Seo, Bumsoo Kim, Daejeong Mun, and Hoonjae Lee (ArtLab) was published in Applied Artificial Intelligence. (*equal contribution)
(2025.01) [IC] "Reflexive Guidance: Improving OoDD in Vision-Language Models via Self-Guided Image-Adaptive Concept Generation" with Jihyo Kim* and Seulbi Lee* was accepted to ICLR 2025 and will be presented in Singapore in April 2025. (*equal contribution)
(2024.09) [Welcome] Minhyeong Lee joined our lab as an M.S. student.
(2024.08) [IJ] "StochCA: A novel approach for exploiting pretrained models with cross-attention" with Seungwon Seo and Suho Lee was published in Neural Networks.
(2024.08) [IJ] "Self-Supervised Representation Learning for Basecalling Nanopore Sequencing Data" with Carlos Vintimilla was published in IEEE Access.
(2024.07) [IJ] "Context-aware Cross Feature Attentive Network for Click-Through Rate Predictions" with Sujin Lee was published in Applied Intelligence.
(2024.06) [DC] "Uncovering Hidden Vulnerabilities in Machine Unlearning: Adversarial Attack as a Probe and Pruning as a Solution" with Huiyeong Lee was presented at 한국정보과학회 2024 한국컴퓨터종합학술대회.
(2024.06) [DC] "균일성 감소 정규화: OODD 성능 향상을 위한 다중모달 미세조정" with Jeonghyeon Kim was presented at 한국정보과학회 2024 한국컴퓨터종합학술대회.
(2024.06) [DC] "안면 피부 상태 평가를 위한 맞춤형 Large Vision Language Model 개발" with Beomsu Kim, Jihyo Kim, Minjun Kang, Huiyeong Lee was presented at 한국정보과학회 2024 한국컴퓨터종합학술대회.
(2024.06) [Welcome] Chanyoung Park joined our lab as a Ph.D student.
(2024.06) [Welcome] Shinhye Lee joined our lab as an M.S. student.
(2024.06) [Welcome] Jongeon Baek joined our lab as an intern.
(2024.05) [Project] "거대 기초 모델의 AI 안정성을 향상시키기 위한 학습 및 지식삭제 알고리즘 연구"가 한국연구재단 중견연구 과제로 선정되었습니다.
(2024.04) [Visiting] Jihyo Kim is currently a visiting student at the University of Sydney, Australia.
(2024.04) [IJ] "Generalized Outlier Exposure: Towards a trustworthy out-of-distribution detector without sacrificing accuracy" with Jiin Koo and Sungjoon Choi was published in Neurocomputing.
(2024.03) [Welcome] Seulbi Lee and Uichan Lee joined our lab as M.S students.
(2024.02) [IC] "Comparison of Out-of-Distribution Detection Performance of CLIP-based Fine-Tuning Methods" with Jeonghyeon Kim and Jihyo Kim was presented at ICEIC 2024. (Best Paper Award)
(2024.02) [Welcome] Deokyong Lee joined our lab as an intern.
(2023.12) [IJ] "A unified benchmark for the unknown detection capability of deep neural networks" with Jihyo Kim and Jiin Koo was published in Expert Systems with Applications.
(2023.08) [Award] Seulbi Lee, Minjun Kang, Inseo Hwang won an award of "우수상" at "인공지능 SCI 부트캠프".
(2023.08) [IJ] "Patch-Level Consistency Regularization in Self-Supervised Transfer Learning for Fine-Grained Image Recognition" with Yejin Lee and Suho Lee was published in Applied Sciences.
(2023.07) [Welcome] Huiyeong Lee joined our lab as intern.
(2023.06) [DC] "시간 정보의 명시적인 반영을 통한 LTSF-Linear 모델의 예측 성능 개선" with Seulbi Lee and Seungwon Seo was presented at 한국정보과학회 2023 한국컴퓨터종합학술대회. (학부생논문경진대회 장려상)
(2023.06) [DC] "Cross-Attention을 활용한 새로운 전이 학습 기법" with Seungwon Seo and Suho Lee was presented at 한국정보과학회 2023 한국컴퓨터종합학술대회. (우수논문상)
(2023.06) [DC] "기준이 불명확한 분류 문제에 적합한 레이블링 방법: 쌍별 비교기반의 순위 레이블링" with Beomsu Kim, Jihyo Kim, Minjun Kang and Daejung Moon was presented at 한국정보과학회 2023 한국컴퓨터종합학술대회.
(2023.06) [DC] "임의 텍스트 미세조정 학습을 통한 CLIP 모델의 학습 외 분포 데이터 탐지 성능 향상 방법" with Jeonghyeon Kim and Jihyo Kim was presented at 한국정보과학회 2023 한국컴퓨터종합학술대회. (우수논문상)
(2023.05) [DJ] "EDAD: 도메인 적응과 지식 증류를 통합한 효율적 도메인 적응 증류" with Seungwon Seo was published in 대한산업공학회지.
(2023.03) [Welcome] Jeonghyeon Kim joined our lab as an M.S / Ph.D integrated student.
(2023.03) [Welcome] Beomsu Kim joined our lab as an M.S. student.
(2023.03) [Welcome] Minjun Kang joined our lab as an intern.
(2023.02) [IC] "Deep Active Learning with Contrastive Learning Under Realistic Data Pool Assumptions" with Jihyo Kim and Jeonghyeon Kim was accepted to AAAI 2023 Workshop on Practical Deep Learning in the Wild as a poster presentation.
(2022.11) [DC] "클릭률 예측을 위한 컨텍스트 기반의 교차 어텐션 추천 시스템" with Sujin Lee was presented at 대한산업공학회 추계학술대회.
(2022.11) [DC] "자기 지도 학습과 메타 정보를 활용한 피부 상태 판별 연구" with Sungchul Kim, Daejung Moon, Beomsu Kim, Seungwon Seo and Jihyo Kim was presented at 대한산업공학회 추계학술대회.
(2022.11) [DC] "Self-supervised Representation Learning for Basecalling Oxford Nanopore Sequencing Data" with Carlos Vintimilla was presented at 대한산업공학회 추계학술대회.
(2022.09) [IC] "Intra-class Contrastive Learning Improves Computer Aided Diagnosis of Breast Cancer in Mammography" with Suho Lee* was accepted to MICCAI 2022 as an oral presentation. (*Work done during an internship at Lunit)
(2022.07) [Welcome] Jeonghyeon Kim joined our lab as intern.
(2022.07) [Competition] Seungwon Seo, Yejin Lee, Daejung Mun won an award for "2022년 인공지능 온라인 경진대회".
(2022.06) [DC] "현실적인 데이터 풀을 고려한 능동적 학습 방법 비교연구" with Ji Hyo Kim was presented at 한국정보과학회 한국컴퓨터종합학술대회.
(2022.06) [DC] "세밀한 객체 인식을 위한 자기 지도 학습 모델의 전이 학습" with Ye Jin Lee was presented at 한국정보과학회 한국컴퓨터종합학술대회. (우수논문상)
(2022.06) [DC] "EDAD: 자연어 처리에서의 효율적 도메인 적응 증류" with Seung Won Seo was presented at 한국정보과학회 한국컴퓨터종합학술대회.
(2022.05) [Welcome] Beomsu Kim and Inseo Hwang joined our lab as intern.
(2021.12) [Welcome] Seulbi Lee joined our lab as an intern.
(2021.10) [Competition] Seungwon Seo and Sungchul Kim won an award for "iDASH 2021 Secure Genome Analysis Competition".
(2021.10) [IJ] "Supervised Contrastive Embedding for Medical Image Segmentation" with Sangwoo Lee*, Yejin Lee*, and Geongyu Lee was published in IEEE Access. (*equal contribution)
(2021.09) [IJ] "Similarity-Based Deep Neural Networks" with Seungyeon Lee, Eunji Jo, Gyeong Bok Jung, Dohyun Kim was published in International Journal of Fuzzy Logic and Intelligent Systems.
(2021.09) [Welcome] Carlos Vintimilla, Soojin Lee, and joined our lab as an M.S. student.
(2021.09) [Welcome] Yongtaek Gong joined our lab as a Ph.D. student.
(2021.07) [IC] "Self-Knowledge Distillation with Progressive Refinement of Targets" with Kyungyul Kim, ByeongMoon Ji, and Doyoung Yoon was accepted to ICCV 2021 as an oral presentation.
(2021.06) [DC] "의료 영상 분할에서의 도메인 일반화를 위한 형태 정보의 활용" with Jun Ho Bae was presented at 대한산업공학회 춘계공동학술대회.
(2021.06) [DC] "딥러닝의 예측 강건성 측면에서의 모델 경량화 효과 연구" with Chanhee Choi was presented at 대한산업공학회 춘계공동학술대회.
(2021.06) [DC] "특정 도메인을 위한 질의응답 시스템의 불확실성 회피 능력에 대한 고찰" with Seungmoo Yang was presented at 대한산업공학회 춘계공동학술대회.
(2021.06) [DC] "Mixup을 활용한 이상 입력 탐지 모델의 불확실성 추정 능력 개선" with Jiin Koo was presented at 대한산업공학회 춘계공동학술대회.
(2021.05) [IJ] "A Unified Defect Pattern Analysis of Wafer Maps Using Density-Based Clustering" with Jaehoon Koo was published in IEEE Access.
(2021.05) [IJ] "Exploiting Global Structure Information to Improve Medical Image Segmentation" with Jaemoon Hwang was published in Sensors.
(2021.04) [DJ] "Improving Domain Generalization Performance for Medical Image Segmentation by Self-Supervised Learning" with Yejin Lee and Sangwoo Lee was published in Journal of the Korean Institute of Industrial Engineers.
(2021.03) [Welcome] Sangwoo Lee and Sungchul Kim joined our lab as an M.S. student.
(2021.02) [Welcome] DaeJeong Mun joined our lab as an intern.
(2021.02) [IJ] "A New Smart Smudge Attack Using CNN" with Hansub Shin, Sungyong Sim, Hyukyoon Kwon, and Younho Lee was published in International Journal of Information Security.
(2020.11) [DC] "영역 분할 모델의 성능 향상을 위한 대조적 손실함수의 활용" with Geongyu Lee was presented at 대한산업공학회 추계학술대회.
(2020.09) [IC] "A Comparison of the Performance of Deep Learning Models for Electric Load Forecasting" with Jiin Koo and Seungmoo Yang was presented at ICESI 2020.
(2020.09) [Welcome] Jihyo Kim joined our lab as a Ph.D. student.
(2020.09) [Welcome] Jun Ho Bae and Yejin Lee joined our lab as an M.S. student.
(2020.07) [Welcome] Seungwon Seo, Sungchul Kim, and Sungrea Hong joined our lab as an intern.
(2020.06) [IC] "Confidence-Aware Learning for Deep Neural Networks" with Jooyoung Moon*, Jihyo Kim*, and Younghak Shin was accepted to ICML 2020. (*equal contribution)
(2020.06) [IJ] "Additive Ensemble Neural Networks" with Minyoung Park, Seungyeon Lee, and Dohyun Kim was published in IEEE Access.
(2020.06) [DJ] "BERT-based Classification Model for Korean Documents" with Dohyun Kim was published in Journal of Society for e-Business Studies.
(2020.04) [IJ] "A New Splitting Criterion for Better Interpretable Trees" with Hyeon Gyu Yeo and Jung-Sik Hong was published in IEEE Access.
(2020.03) [Welcome] Jiin Koo joined our lab as an M.S. student.
(2019.11) [DC] "해부학적 구조를 반영한 흉부 X-ray 영상에서의 폐 영역 분할 모델" with Jaemoon Hwang was presented at 대한산업공학회 추계학술대회.
(2019.11) [DC] "심층 신경망의 과한 확신을 방지하는 새로운 정규화 방법" with Jooyoung Moon and Jihyo Kim was presented at 대한산업공학회 추계학술대회.
(2019.09) [Welcome] Chanhee Choi and Seungmoo Yang joined our lab as an M.S. student.
(2019.07) [Welcome] Jinsung Jeon, Sanghyeon Kim, Jiin Koo, Sangwoo Lee, and Yejin Lee joined our lab as an intern.
(2019.03) [Welcome] Geongyu Lee joined our lab as an M.S. student.
(2018.09) [Welcome] Jihyo Kim joined our lab as an M.S. student.
(2018.03) [Welcome] Jaemoon Hwang and Jooyoung Moon joined our lab as an M.S. student.
Page updated
Report abuse