Seoul National University · ECE · IPAI
Machine
Intelligence Lab
We research natural language processing, deep learning, and AI applications including LLMs, multimodal learning, and trustworthy AI under Prof. Kyomin Jung.
Updates
Latest News
-
08/2026연구실 김민범 학생 2026년 8월 협동과정 인공지능전공 박사 졸업 최우수논문상 (Best Ph.D Dissertation Award) 수상.
-
08/2026Our 9 papers are accepted to EMNLP 2026.
- Hyukhun Koh et al., "When i.i.d. Factorization Breaks Language: The Structural Failure of Discrete Diffusion Language Models" is accepted to EMNLP 2026.
- Jiwon Moon et al., "Language Shapes Instruction Hierarchy Compliance in Multilingual LLMs" is accepted to EMNLP 2026.
- Yunah Jang et al., "How You Ask Matters! Adaptive RAG Robustness to Query Variations" is accepted to EMNLP 2026.
- Yerin Hwang et al., "How Instruction Hierarchy Breaks Under Pairwise Multi-Conflict Pressure" is accepted to EMNLP 2026.
- Sangmook Lee et al., "Are We Measuring Strategy or Phrasing? The Gap Between Surface- and Approach-Level Diversity in LLM Math Reasoning" is accepted to EMNLP 2026.
- Hyojeong Yu et al., "PRAGMA: Evaluating Personalized Guidance with Memory Alignment in Lifelong Conversations" is accepted to EMNLP 2026.
- Yunah Jang et al., "CompCQR: Compositional Query Generation for Training-Free Conversational Search" is accepted to Findings of EMNLP 2026.
- Kyeongman Park et al., "A Benchmark for the Generation and Evaluation of Scientific Architecture Diagrams in HTML/SVG" is accepted to Findings of EMNLP 2026.
- Junseok Kim et al., "Personalized Privacy Control in LLMs via Attention Head Intervention" is accepted to Findings of EMNLP 2026.
-
08/2026Junsu Park et al., "Instructional Video 기반 멀티모달 절차 계획 생성 방식 비교" KCC 2026 우수발표논문상 수상
-
08/2026연구실 졸업생 황예린 박사(2026년 8월 졸업), 2026년 9월부터 동국대학교 컴퓨터-AI학부 교수 임용.
-
08/2026연구실 졸업생 김용일 박사(2024년 8월 졸업), 2026년 9월부터 가톨릭대학교 인공지능학과 교수 임용.
-
06/2026
Our paper "ReflectCAP: Detailed Image Captioning with Reflective Memory" is accepted to ECCV 2026.
-
05/2026Our 3 papers are accepted to ICML 2026.
- Our paper "CausalArmor: Efficient Indirect Prompt Injection Guardrails via Causal Attribution" is accepted to ICML 2026.
- Our paper "Beyond Normalization: Rethinking the Partition Function as a Difficulty Scheduler for RLVR" is accepted to ICML 2026.
- Our paper "PEARL: Differentially Private and Entropy-Aware Regulated Language Generation" is accepted to ICML 2026.
-
04/2026Our 5 papers are accepted to ACL 2026.
- Our paper "How Training Data Shapes the Use of Parametric and In-Context Knowledge in Language Models" is accepted to ACL 2026.
- Our paper "Evaluating Visual Narrative Coherence in Story Visualization via Diversified Storylines" is accepted to ACL 2026 as an Oral presentation.
- Our paper "Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning" is accepted to ACL 2026 Findings.
- Our paper "Rethinking Post-Unlearning Behavior of Large Vision-Language Models" is accepted to ACL 2026 Findings.
- Our paper "A Universal Avoidance Method for Diverse Multi-branch Generation" is accepted to ACL 2026 Findings.
-
02/2026연구실 양낙영 학생 2026년 2월 협동과정 인공지능전공 박사 졸업 우수논문상 (Distinguished Ph.D Dissertation Award) 수상.
-
01/2026Our paper "Erase or Hide? Suppressing Spurious Unlearning Neurons for Robust Unlearning" is accepted to ICLR 2026.
-
01/2026Our 2 papers are accepted to EACL 2026.
- Our paper "Don't Judge Code by Its Cover: Exploring Biases in LLM Judges for Code Evaluation" is accepted to EACL 2026 Findings.
- Our paper "Persona Switch: Mixing Distinct Perspectives in Decoding Time" is accepted to EACL 2026 Findings.
-
01/2026Our paper "Confidence-Guided Stepwise Model Routing for Cost-Efficient Reasoning" is accepted to AAAI 2026.
-
09/2025Our paper "Program Synthesis via Test-Time Transduction" is accepted to NeurIPS 2025.
-
08/2025Our 9 papers are accepted to EMNLP 2025.
- Our paper "Conditional [MASK] Discrete Diffusion Language Model" is accepted to EMNLP 2025.
- Our paper "Avoidance Decoding for Diverse Multi-Branch Story Generation" is accepted to EMNLP 2025.
- Our paper "Fooling the LVLM Judges: Visual Biases in LVLM-Based Evaluation" is accepted to EMNLP 2025.
- Our paper "FaithUn: Toward Faithful Forgetting in Language Models by Investigating the Interconnectedness of Knowledge" is accepted to EMNLP 2025.
- Our paper "Harmful Prompt Laundering: Jailbreaking LLMs with Abductive Styles and Symbolic Encoding" is accepted to EMNLP 2025.
- Our paper "ReflAct: World-Grounded Decision Making in LLM Agents via Goal-State Reflection" is accepted to EMNLP 2025.
- Our paper "Public Data Assisted Differentially Private In-Context Learning" is accepted to EMNLP 2025 Findings.
- Our paper "Can You Trick the Grader? Adversarial Persuasion of LLM Judges" is accepted to EMNLP 2025 Findings.
- Our paper "Drift: Decoding-time Personalized Alignments with Implicit User Preferences" is accepted to EMNLP 2025 Findings.
-
08/2025Our paper "Unplug and Play Language Models: Decomposing Experts in Language Models at Inference Time" is accepted to CIKM 2025.
-
06/2025Our 2 papers are accepted to ACL 2025.
- Our paper "LLMs can be easily Confused by Instructional Distractions" is accepted to ACL 2025.
- Our paper "A Character-Centric Creative Story Generation via Imagination" is accepted to ACL 2025.
-
01/2025Our paper "Guaranteed Generation from Large Language Models" is accepted to ICLR 2025.
-
01/2025Our 6 papers are accepted to NAACL 2025.
- Our paper "Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation" is accepted to NAACL 2025 as an Oral presentation.
- Our paper "Generating Diverse Hypotheses for Inductive Reasoning" is accepted to NAACL 2025 as an Oral presentation.
- Our paper "AdvisorQA: Towards Helpful and Harmless Advice-seeking Question Answering with Collective Intelligence" is accepted to NAACL 2025.
- Our paper "SWITCH: Studying with Teacher for Knowledge Distillation of Large Language Models" is accepted to NAACL 2025 Findings.
- Our paper "VLind-Bench: Measuring Language Priors in Large Vision-Language Models" is accepted to NAACL 2025 Findings.
- Our paper "Mitigating Hallucinations in Large Vision-Language Models via Summary-Guided Decoding" is accepted to NAACL 2025 Findings.
-
12/2024Our paper "Return of EM: Entity-driven Answer Set Expansion for QA Evaluation" is accepted to COLING 2025 as an Oral presentation.
-
09/2024Our 2 papers are accepted to EMNLP 2024.
- Our paper "MP2D: An Automated Topic Shift Dialogue Generation Framework Leveraging Knowledge Graphs" is accepted to EMNLP 2024.
- Our paper "Can LLMs Recognize Toxicity? A Structured Investigation Framework and Toxicity Metric" is accepted to EMNLP 2024 Findings.
-
08/2024Our paper "MVMR: A New Framework for Evaluating Faithfulness of Video Moment Retrieval against Multiple Distractors" is accepted to CIKM 2024.
-
08/2024Our paper "Mitigating Biases for Instruction-following Language Models via Bias Neurons Elimination" is accepted to ACL 2024.
-
06/2024Our 3 papers are accepted to NAACL 2024.
- Our paper "LifeTox: Unveiling Implicit Toxicity in Life Advice" is accepted to NAACL 2024 as an Oral presentation.
- Our paper "Fine-grained Gender Control in Machine Translation with Large Language Models" is accepted to NAACL 2024.
- Our paper "IterCQR: Iterative Conversational Query Reformulation with Retrieval Guidance" is accepted to NAACL 2024.
-
12/2023Our paper "LongStory: Coherent, Complete and Length Controlled Long story Generation" from our new member is accepted to PAKDD 2024 as an Oral presentation.
-
12/2023Our paper "Boosting Speech Enhancement with Clean Self-Supervised Features via Conditional Variational Autoencoders" is accepted to ICASSP 2024.
-
12/2023Our 6 papers are accepted to EMNLP 2023.
- Our paper "DPP-TTS: Diversifying prosodic features of speech via determinantal point processes" is accepted to EMNLP 2023.
- Our paper "Weakly Supervised Semantic Parsing with Execution-based Spurious Program Filtering" is accepted to EMNLP 2023.
- Our paper "Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation" is accepted to EMNLP 2023.
- Our paper "Dialogizer: Context-aware Conversational-QA Dataset Generation from Textual SourcesA" is accepted to EMNLP 2023.
- Our paper "Asking Clarification Questions to Handle Ambiguity in Open-Domain QA" is accepted to EMNLP 2023 Findings.
- Our paper "PR-MCS: Perturbation Robust Metric for MultiLingual Image Captioning" is accepted to EMNLP 2023 Findings.
-
10/2023정교민 교수님 공과대학 2023년도 신양공학학술상 수상
-
08/2023Our paper "Leveraging Ensemble Techniques and Metadata for Subjective Knowledge-grounded Conversational Systems" is awarded as the Best paper in DSTC11 workshop.
-
05/2023Our 3 papers are accepted to ACL 2023.
- Our paper "BREAK: Breaking the Dialogue State Tracking Barrier with Beam Search and Re-ranking" is accepted to ACL 2023.
- Our paper "Critic-Guided Decoding for Controlled Text Generation" is accepted to ACL 2023 Findings.
- Our paper "Injecting Comparison Skills in Task-Oriented Dialogue Systems for Database Search Results Disambiguation" is accepted to ACL 2023 Findings.
-
04/2023Seongho Joo, Kangil Lee, Kyungmin Min ranked first in the response accuracy category in DSTC11 challenge.
-
03/2023연구실 졸업생 이환희 박사 중앙대학교 AI학과 교수 임용.
-
01/2023Our paper "Task-specific Compression for Multi-task Language Models using Attribution-based Pruning" is accepted to EACL 2023 Findings.
-
08/2022연구실 이환희 학생 박사 졸업 우수논문상 (Distinguished Ph.D Dissertation Award) 수상.
-
07/2022연구실 이윤형 학생 율촌 AI STAR 장학생 선정.
-
04/2022Our paper "Masked Summarization to Generate Factually Inconsistent Summaries for Improved Factual Consistency Checking" is accepted to NAACL 2022 Findings.
-
04/2022Our paper "Modality Alignment between Deep Representations for Effective Video-and-Language Learning" is accepted to LREC 2022.
-
02/2022We ranked the first in publication performance among the BK21 computer module labs! (2020.02.01 - 2022.1.31).
-
01/2022Three papers "VarianceFlow: High-quality and Controllable Text-to-Speech Using Variance Information via Normalizing Flow", "Deriving Explainable Discriminative Attributes using Confusion about Counterfactual Class" and "Subgraph Representation Learning With Hard Negative Samples for Inductive Link Prediction" are accepted to ICASSP 2022.
-
12/2021Prof. Jung receives the 2021 IEEE ICDM 10-Year Highest-Impact Paper Award for the paper "IRIE: Scalable and Robust Influence Maximization in Social Networks".
-
09/2021Our paper "Contrastive Learning for Context-aware Neural Machine Translation Using Coreference Information" is accepted to WMT 2021.
-
09/2021Prof. Jung is serving as a vice-Chair of the ECE department for student affairs.
-
09/2021Prof. Jung receives an excellent lecture award (우수강의상) for the Programming Methodology class.
-
09/2021Prof. Jung is promoted to a full professor with tenure.
-
08/2021연구실 이환희 학생 율촌 AI STAR 장학생 선정.
-
08/2021Our paper "QACE: Asking Questions to Evaluate an Image Caption" is accepted to EMNLP 2021 Findings.
-
08/2021Two papers "CrossAug: A Contrastive Data Augmentation Method for Debiasing Fact Verification Models" and "RABERT: Relation-Aware BERT for Target-Oriented Opinion Words Extraction" are accepted to CIKM 2021.
-
06/2021Our paper "모달리티 정렬을 통한 비디오 질의 응답 시스템 개선" is awarded as the Best Paper (최우수논문상) in KCC 2021.
-
05/2021Our paper "Interpretable Disease Prediction Using Heterogeneous Patient Records with Self-Attentive Fusion Encoder" is accepted to Journal of American Medical Informatics Association.
-
05/2021Our paper "UMIC: An Unreferenced Metric for Image Captioning via Contrastive Learning" is accepted to ACL 2021.
-
04/2021Our paper "Entangled Bidirectional Encoder to Autoregressive Decoder for Sequential Recommendation" is accepted to SIGIR 2021.
-
03/2021Our paper "KPQA: A Metric for Generative Question Answering Using Keyphrase Weights" is accepted to NAACL 2021.
-
01/2021Our paper "Bidirectional Variational Inference for Non-Autoregressive Text-to-Speech" is accepted to ICLR 2021.
-
12/2020Our paper "Neural Sequence-to-grid Module for Learning Symbolic Rules" is accepted to AAAI 2021.
-
08/2020연구실 윤승현 학생 박사 졸업 우수논문상 (Distinguished Ph.D Dissertation Award) 수상.
-
07/2020Our paper "Multimodal Speech Emotion Recognition using Cross Attention with Aligned Audio and Text" is accepted to Interspeech 2020.
-
05/2020Our paper "Comparative Studies on Machine Learning for Paralinguistic Signal Compression and Classification" is accepted for publication in the Journal of Supercomputing. (SCI, IF=2.157).
-
04/2020Our paper "Fast and Accurate Deep Bidirectional Language Representations for Unsupervised Learning" is accepted to ACL 2020.
-
02/2020Our paper "맥락 정보 유사도를 활용한 이미지 피처 해싱" is selected as the 우수발표논문상 in KSC 2019.
-
02/2020Our paper "Propagate-Selector: Detecting Supporting Sentences for Question Answering via Graph Neural Networks" is accepted to LREC 2020.
-
01/2020Our papers "Drug-disease Graph: Predicting Adverse Drug Reaction Signals via Graph Neural Network with Clinical Data" and "Reliable Aggregation Method for Vector Regression Tasks in Crowdsourcing" are accepted to PAKDD 2020.
-
01/2020Our paper "Attentive Modality Hopping Mechanism for Speech Emotion Recognition" is accepted to ICASSP 2020 as oral presentation.
-
11/2019Our patent "인공 지능 기반의 대화 시스템 및 그 응답 제어 방법, No 10-2059015" is registered on 2019. 12., Korea.
-
11/2019Our paper "Improving Context-Aware Neural Machine Translation Using Self-Attentive Sentence Embedding" is accepted to AAAI 2020.
-
11/2019Our paper "Efficient spread-size approximation of opinion spreading in general social networks" is accepted to Physical Review E.
-
09/2019Our paper "Effective Sentence Scoring Method Using BERT for Speech Recognition" is accepted to ACML 2019.
-
09/2019Our paper "A Compare-Aggregate Model with Latent Clustering for Answer Selection" is accepted to CIKM 2019 as oral presentation.
-
02/2019Our papers "문장 레벨 그래프 회선 신경망을 통한 텍스트 분류" and "이미지 물체 위치탐색을 위한 효율적인 크라우드소싱 시스템" are selected as the 우수발표논문상 in KSC 2018.
-
02/2019Our paper "Speech Emotion Recognition Using Multi-Hop Attention Mechanism" is accepted ICASSP 2019 as oral presentation.
-
01/2019Etnews reported our research result: 서울대,카이스트, '낚시성 기사' 잡아내는 딥러닝 모델 기술 개발
-
11/2018Our paper "Neural networks for compressing and classifying speaker-independent paralinguistic signals" is accepted to IEEE BigComp 2019.
-
11/2018Thanks to past collaborative research activities with Adobe, our lab has been selected for Adobe Research Software Donation Program.
-
11/2018Three papers "Improving Neural Question Generation using Answer Separation", "Detecting Incongruity Between News Headline and Body Text via a Deep Hierarchical Encoder" (with KAIST Data Science lab), "Number Sequence Prediction Problems and Computational Powers of Neural Network Models", are accetped at AAAI 2019!
-
07/2018Our paper "Mobile App Recommendation with Sequential App Usage Behavior Tracking" is accepted to Journal of Internet Technology.
-
02/2018Our paper "Learning to Rank Question-Answer Pairs using Hierarchical Recurrent Encoder with Latent Topic Clustering" is accepted to NAACL 2018.
-
12/2017Our paper "Multi-scale Nystrom Method" is accepted to AISTATS 2018 as oral presentation.
-
12/2017Our research group took 5th place in AI R&D Fake News Challenge.
-
11/2017Our paper "Contextual-CNN: A Novel Architecture Capturing Unified Meaning for Sentence Classification" is accetped to IEEE BigComp 2018.
-
09/2017Our paper "Synonym Discovery with Etymology-based Word Embeddings" is accepted to IEEE SSCI 2017.
-
05/2017Our paper "Hierarchical Ordering with Partial Pairwise Hierarchical Relationships on the Marcaque Brain Data Sets" is accepted to PLOS ONE.
-
06/2017Prof. Jung had an interview with KBS about rumor detection for AI.
-
06/2017Prof. Jung gave interviews to tbs and ebs about our research on rumor detectionin social media.
-
11/2016Our paper "Rumor Detection over Varying Time Windows" is accepted to PLOS ONE.
-
07/2016Our paper "Knowledge extraction through etymological networks: Synonym discovery in sino-korean words" is selected as the best paper of the information technology session in ICIKM 2016.
-
06/2016한국연구재단 중견연구자지원사업 선정(의료 빅데이터 기반 약물의 미발견 위해정보 검출 기계학습 기법 개발)
-
06/2016Our paper "Knowledge extraction through etymological networks: Synonym discovery in sino-korean words" is accepted to ICIKM 2016.
-
05/2016Our paper "Phase transitions for information diffeusion in random clustered networks" is accepted to European Physical Jounral B.
-
05/2016서울대학교 공학연구원 한우물로 홈런치기 과제 선정(Learning 기반 한국어 자연어 처리 시스템 및 영-한 기계번역 시스템 개발)
-
04/2016Our paper "Automatic Question Answering System for Consumer Products" is accepted to IntelliSys 2016.
-
03/2016정교민 교수님 부교수 승진
-
02/2016Prof. Jung gave a talk "Deep learning for Natural Language Processing" at Facebook, Menlo Park, CA, USA.
-
12/2015정교민 교수님 자연과학대학 수리과학부 겸무교수로 위촉
-
12/2015Our paper "Centrality Fairness: Measuring and Analyzing Structual Inequality of Online Social Network" is accepted to JIT.
-
08/2015연구실 김정명 학생 석사 졸업 우수논문상 (Distinguished Dissertation Award) 수상.
-
05/2015Prof. Jung gave a Tutorial Talk "Deep Learning for the Web" at the WWW 2015.
-
05/2015Our paper "Minimizing Expected Losses in Perturbation Models with Multidimensional Parametric Min-cuts" is accepted to UAI 2015 as a plenary presentation.
-
05/2015Our paper "Double Nystrom Method: An Efficient and Accurate Nystrom Scheme for Large-Scale Data Sets" is accepted to ICML 2015.
-
03/2015Our paper "Reliable Multiple-choice Iterative Algorithm for Crowdsourcing Systems" is accepted to SIGMETRICS 2015.
-
01/2014트위터 등 SNS에서의 루머를 90% 정도의 정확도로 가려내는 시스템 개발 - KBS1 뉴스라인 대담, 아리랑 TV, tbs eFM 출연
Research
Recent Publications
-
When i.i.d. Factorization Breaks Language: The Structural Failure of Discrete Diffusion Language Models
Hyukhun Koh, Seongho Joo, Haesung Pyun, Kyomin JungEMNLP 2026 August 2026 · Budapest, Hungary -
Language Shapes Instruction Hierarchy Compliance in Multilingual LLMs
Jiwon Moon, Yerin Hwang, Kyomin JungEMNLP 2026 August 2026 · Budapest, Hungary -
How You Ask Matters! Adaptive RAG Robustness to Query Variations
Yunah Jang, Megha Sundriyal, Kyomin Jung, Meeyoung ChaEMNLP 2026 August 2026 · Budapest, Hungary -
How Instruction Hierarchy Breaks Under Pairwise Multi-Conflict Pressure
Yerin Hwang, Jiwon Moon, Changho Lee, Kwangmin Ki, taegwan kang, Jinsik Lee, Honglak Lee, Kyomin JungEMNLP 2026 August 2026 · Budapest, Hungary -
Are We Measuring Strategy or Phrasing? The Gap Between Surface- and Approach-Level Diversity in LLM Math Reasoning
Sangmook Lee, Minbeom Kim, Jeonghye Kim, Dohyung Kim, Sojeong Rhee, Kyomin JungEMNLP 2026 August 2026 · Budapest, Hungary