Knowledge Editing
I develop methods for selectively modifying the knowledge stored in large models, spanning machine unlearning, continual learning, and modular knowledge representations.
Associate Professor
Junsuk Choe
Multimodal foundation models · Knowledge editing · Grounding
Department of Computer Science and Engineering, Sogang University
Office: AS-913 · Phone: +82-2-705-8495
I am an associate professor in the Department of Computer Science and Engineering at Sogang University. Before joining Sogang University in Fall 2021, I spent two wonderful years as a Research Intern and then a Research Scientist at NAVER AI Lab. I received my Ph.D. from Yonsei University in 2020 under the supervision of Prof. Hyunjung Shim, and my B.S. from Yonsei University in 2013.
Recruiting One new lab member.
Updates
News
- Minyoung will visit Johns Hopkins University as a visiting researcher, hosted by Prof. Jaemin Cho.
- Jimin will visit the University of British Columbia as a visiting researcher, hosted by Prof. Cristina Conati.
- We have 9 papers: 1 AAAI, 1 ICML, 1 COLM, 1 NeurIPS, 4 Pattern Recognition Letters, and 1 Neurocomputing.
- Promoted to Associate Professor at Sogang University. Thanks to all my students, colleagues, and collaborators for their support.
- Yeji joined NAVER AI Lab as a research intern.
- Granted the 생성AI 선도인재양성사업 as Co-PI, led by NC AI and funded by IITP, and the 독자 AI 파운데이션 모델 프로젝트 as Co-PI, led by Upstage and funded by NIPA/NIA/IITP.
- Dongjun visited the University of Tübingen as a visiting researcher, hosted by Prof. Seong Joon Oh.
- Minyoung, Yeji, and Yejin visited the University of Tübingen as visiting researchers, hosted by Prof. Seong Joon Oh.
- Hyo Seo visited Michigan State University as a visiting researcher, hosted by Prof. Sijia Liu.
- Granted the Global Humanities and Social Convergence Research Program (1.8B KRW) as Co-PI from NRF Korea, led by Prof. Dasaem Jeong.
- Dongjun participated in the IITP AI Intensive Education Program at Carnegie Mellon University.
- Granted the Outstanding Young Scientist Grant (900M KRW) as PI from NRF Korea, with Prof. Buru Chang as Co-PI.
- Minyoung and Hyo Seo visited the University of Tübingen as visiting researchers, hosted by Prof. Seong Joon Oh and Prof. Zeynep Akata.
- Joined the Department of Computer Science and Engineering at Sogang University as Assistant Professor.
- Defended Ph.D. thesis and joined NAVER AI Lab as Research Scientist.
Focus
Research Interests
My research centers on large multimodal foundation models, including vision-language and video-language models, and aims to make them sustainable to manage and reliable to deploy without costly retraining. I pursue this goal along two complementary directions: knowledge editing and grounding.
Grounding
I work on keeping multimodal model outputs faithful to the intended visual and textual evidence, addressing hallucination, cross-image leakage, and cross-modal interaction imbalance.
Lab
People
Current Members
8- Dongjun Hwang
- Minyoung Lee
- Yeji Park
- Beomyun Kwon
- Jimin Hong
- Elena Kuular
- Jihong Park
- Seungjun Ha
Alumni
4Service
Academic Activities
Area Chair
- COLM 2025-26
- ICLR 2026
- NeurIPS 2026
Reviewer
- Computer Vision: CVPR, ICCV, ECCV, WACV, ACCV
- Machine Learning / AI: ICML, ICLR, NeurIPS, CoLLAs, AAAI, AISTATS, IJCAI
- Journals: IEEE PAMI, IJCV, IEEE TIP, IEEE TCSVT
Output
Publications
J/C/W/P denote journal, conference, workshop, and preprint. * co-first, ^ co-corresponding.
Enhancing Cross-Modal Interactions in Audio-Visual LLMs via Training-Free Attention Injection
Minyoung Lee, Jimin Hong, Junsuk Choe
Under review, 2026.
Inserting Random Tokens at the Tail: A Simple Training-Free Input Perturbation for LVLMs
Jimin Hong, Minyoung Lee, Junsuk Choe
Under review, 2026.
Mitigating Cross-Image Information Leakage in Multi-Image Understanding with Large Vision-Language Models
Yeji Park, Minyoung Lee, Sanghyuk Chun, Junsuk Choe
Under review, 2026.
Knowledge Vector Weakening: Efficient Training-free Unlearning for Large Vision-Language Models
Yejin Kim, Dongjun Hwang, Sungmin Cha, Junsuk Choe
Under review, 2026.
ALISE: Alignment-aware Data Selection for Unlearning in Contrastive Vision-Language Models
Dongjun Hwang, Yejin Kim, Beomyun Kwon, Junsuk Choe
Under review, 2026. Extended version of the ICML 2026 MemFM workshop paper.
Alignment-aware Data Selection for Unlearning in Contrastive Vision-Language Models
Dongjun Hwang, Yejin Kim, Beomyun Kwon, Junsuk Choe
ICML 2026 Workshop on the Impact of Memorization on Trustworthy Foundation Models.
Behavioral Proxy Conditioning for Financial Stress Scenario Generation with a Pretrained Diffusion Model
Elena Kuular, Junsuk Choe
ICML 2026 Workshop on Foundation Models for Structured Data.
EvoVLM: Multimodal Evolutionary Feedback for Visual Symbolic Regression
Hyejin Lee, Junsuk Choe
ICML 2026 Workshop on AI for Science.
Rarr: Real-Time Attention-Driven Rain Removal With Hierarchical Scale-Aware Efficient Network
Seungho Eum, Ihjjoon Cho, Jeonghyeon Kim, Junsuk Choe, Unsang Park
ICPR 2026.
Training-Free Uncertainty-guided Logit Adjustment for Few-Shot Class-Incremental Learning
Sungwon Woo, Dongjun Hwang, Shiwon Kim, Junsuk Choe, Jongho Nang
CVPR 2026 Findings.
Grounding the "Not": Symbolic Representation of Negation for Logical Reasoning in VLMs
Inha Kang, Seonho Lee, Jiho Choi, Junsuk Choe, Hyunjung Shim
ICLR 2026 Workshop on Logical Reasoning of Large Language Models.
Enhancing Multi-Image Understanding through Delimiter Token Scaling
Minyoung Lee, Yeji Park, Dongjun Hwang, Yejin Kim, Seong Joon Oh, Junsuk Choe
ICLR 2026.
What "Not" to Detect: Negation-Aware VLMs via Structured Reasoning and Token Merging
Inha Kang, Youngsun Lim, Seonho Lee, Jiho Choi, Junsuk Choe, Hyunjung Shim
ICLR 2026.
OVS Meets Continual Learning: Towards Sustainable Open-Vocabulary Segmentation
Dongjun Hwang, Yejin Kim, Minyoung Lee, Seong Joon Oh, Junsuk Choe
NeurIPS 2025.
Improving Fisher Information Estimation and Efficiency for LoRA-based LLM Unlearning
Yejin Kim*, Eunwon Kim*, Buru Chang^, Junsuk Choe^
COLM 2025.
LMLT: Low-to-high Multi-Level Vision Transformer for Lightweight Image Super-Resolution
Jeongsoo Kim, Jongho Nang, Junsuk Choe
ICCV 2025 Workshop on Advances in Image Manipulation.
Small Object Matters in Weakly Supervised Object Localization
Dongjun Hwang, Seong Joon Oh, Junsuk Choe
Neurocomputing, 2025.
NegMerge: Sign-Consensual Weight Merging for Machine Unlearning
Hyo Seo Kim, Dongyoon Han^, Junsuk Choe^
ICML 2025.
VHOIP: Video-based Human-Object Interaction recognition with CLIP Prior knowledge
Doyeol Baek, Junsuk Choe
Pattern Recognition Letters, 2025.
Fog-Free Training for Foggy Scene Understanding
Minyoung Lee, Kyungwoo Song, Junsuk Choe
Pattern Recognition Letters, 2025.
Curriculum Learning with Class-label Composition for Weakly Supervised Semantic Segmentation
Dongjun Hwang, Hyoseo Kim, Doyeol Baek, Hyunbin Kim, Inhye Kye, Junsuk Choe
Pattern Recognition Letters, 2025.
ConVis: Contrastive Decoding with Hallucination Visualization for Mitigating Hallucinations in Multimodal Large Language Models
Yeji Park*, Deokyeong Lee*, Junsuk Choe^, Buru Chang^
AAAI 2025.
Improving ViT Interpretability with Patch-level Mask Prediction
Junyong Kang, Byeongho Heo, Junsuk Choe
Pattern Recognition Letters, 2025.
Discovering an Inference Recipe for Weakly-Supervised Object Localization
Sanghuk Lee*, Cheolhyun Mun*, Youngjung Uh, Junsuk Choe, Hyeran Byun
Pattern Recognition, 2024.
Weakly-supervised Incremental learning for Semantic segmentation with Class Hierarchy
Hyoseo Kim, Junsuk Choe
Pattern Recognition Letters, 2024.
Weakly Supervised Semantic Segmentation for Driving Scenes
Dongseob Kim*, Seungho Lee*, Junsuk Choe, Hyunjung Shim
AAAI 2024.
Small Objects Matters in Weakly-supervised Semantic Segmentation
Cheolhyun Mun*, Sanghuk Lee*, Youngjung Uh, Junsuk Choe, Hyeran Byun
WACV 2024.
Neglected Free Lunch - Learning Image Classifiers Using Annotation Byproducts
Dongyoon Han*, Junsuk Choe*, Seonghyeok Chun, John Chung, Minsuk Chang, Sangdoo Yun, Jean Song, Seong Joon Oh
ICCV 2023.
Entropy Regularization for Weakly Supervised Object Localization
Dongjun Hwang, Jung-Woo Ha, Hyunjung Shim, Junsuk Choe
Pattern Recognition Letters, 2023.
Evaluation for Weakly Supervised Object Localization: Protocol, Metrics, and Datasets
Junsuk Choe*, Seong Joon Oh*, Sanghyuk Chun, Seungho Lee, Zeynep Akata, Hyunjung Shim
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023.
Weakly Supervised Semantic Segmentation using Out-of-Distribution Data
Jungbeom Lee, Seong Joon Oh, Sangdoo Yun, Junsuk Choe, Eunji Kim, Sungroh Yoon
CVPR 2022.
Attention-based Dropout Layer for Weakly Supervised Single Object Localization and Semantic Segmentation
Junsuk Choe*, Seungho Lee*, Hyunjung Shim
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021.
Keep CALM and Improve Visual Feature Attribution
Jae Myung Kim*, Junsuk Choe*, Zeynep Akata, Seong Joon Oh
ICCV 2021.
Normalization Matters in Weakly Supervised Object Localization
Jeesoo Kim, Junsuk Choe, Sangdoo Yun, Nojun Kwak
ICCV 2021.
Contrastive Attention Maps for Self-supervised Co-localization
Minsong Ki, Youngjung Uh, Junsuk Choe, Hyeran Byun
ICCV 2021.
Rethinking Spatial Dimensions of Vision Transformers
Byeongho Heo, Sangdoo Yun, Dongyoon Han, Sanghyuk Chun, Junsuk Choe, Seong Joon Oh
ICCV 2021.
Region-based Dropout with Attention Prior for Weakly Supervised Object Localization
Junsuk Choe, Dongyoon Han, Sangdoo Yun, Jung-Woo Ha, Seong Joon Oh, Hyunjung Shim
Pattern Recognition, 2021.
Re-labeling ImageNet: from Single to Multi-Labels, from Global to Localized Labels
Sangdoo Yun, Seong Joon Oh, Byeongho Heo, Dongyoon Han, Junsuk Choe, Sanghyuk Chun
CVPR 2021.
Evaluating Weakly Supervised Object Localization Methods Right
Junsuk Choe*, Seong Joon Oh*, Seungho Lee, Sanghyuk Chun, Zeynep Akata, Hyunjung Shim
CVPR 2020.
CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features
Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, Youngjoon Yoo
ICCV 2019. Oral presentation.
Attention-based Dropout Layer for Weakly Supervised Object Localization
Junsuk Choe, Hyunjung Shim
CVPR 2019. Oral presentation.
An Empirical Evaluation on Robustness and Uncertainty of Regularization Methods
Sanghyuk Chun, Seong Joon Oh, Sangdoo Yun, Dongyoon Han, Junsuk Choe, Youngjoon Yoo
ICML 2019 Workshop on Uncertainty & Robustness in Deep Learning.
Robust approach to inverse lighting using RGB-D images
Junsuk Choe, Hyunjung Shim
Information Sciences, 2018.
Face generation for low-shot learning using generative adversarial networks
Junsuk Choe*, Song Park*, Kyungmin Kim*, Joo Hyun Park*, Dongseob Kim*, Hyunjung Shim
ICCV 2017 Workshop on MS-Celeb-1M Challenge. Oral presentation.