Research

Building AI that reasons over
relational structures

The LearnData Lab develops machine learning and data mining methods for complex relational data—graphs, knowledge bases, and multi-modal streams. We pursue two interconnected research thrusts: advancing graph-structured learning with scalable, principled models, and building trustworthy, knowledge-enhanced AI that is explainable, robust, and grounded in structured knowledge. Our work has been published at top venues including ICML, ICLR, KDD, WWW, CVPR, ACL, and AAAI.

50+
Publications
17
Lab Members
2
Core Thrusts
2020–
Est. at SKKU
GRAPH-STRUCTURED LEARNING
Thrust 1

Machine Learning & Data Mining for Graph-Structured Data

We develop novel graph neural network/transformer architectures, self-supervised graph representation methods, and federated graph learning frameworks. Our research addresses fundamental challenges in knowledge graph reasoning, recommendation systems, and node/graph-level representation learning, with methods published at KDD, WWW, ICLR, WSDM, and CIKM. We focus on principled approaches—information-theoretic decomposition, conditional independence, and curriculum-guided optimization—that deliver both strong empirical performance and theoretical grounding.

Graph Neural Networks Knowledge Graph Reasoning Multi-Sensor Learning Federated Graph Learning Recommender Systems Self-supervised Learning Numerical Reasoning
Selected Publications
ICLR 2026 TOP VENUE
Low-pass Personalized Subgraph Federated Recommendation
W. Sim, H. Park*
WWW 2025 TOP VENUE
Balancing Graph Embedding Smoothness in Self-Supervised Learning via Information-Theoretic Decomposition
H. Jung, H. Park*
WSDM 2025 TOP VENUE — Oral (Top 6.5%)
CIMAGE: Exploiting the Conditional Independence in Masked Graph Auto-encoders
J. Park+, H. Jung+, H. Park*
KDD 2023 TOP VENUE
Exploiting Relation-aware Attribute Representation Learning in Knowledge Graph Embedding for Numerical Reasoning
S. Kim+, G. Kim+, K. K. Kim, S. Park, H. Jung, H. Park*
TRUSTWORTHY & KNOWLEDGE-ENHANCED AI
Thrust 2

Trustworthy & Knowledge-Enhanced AI

We build AI systems that are not only powerful but also transparent, robust, and grounded in structured knowledge. Our research develops explainability methods for graph neural networks and other generative models, knowledge-based LLM steering, adversarial robustness techniques, and watermarking methods. We also investigate LLM reasoning, multi-modal question answering, and AI for scientific discovery—pushing AI toward systems humans can understand and trust.

Explainable AI (XAI) Adversarial Robustness Multi-modal LLM Reasoning Multi-agent Learning LLM Orchestration AI for Science Knowledge Graph QA
Selected Publications
CVPR 2026 TOP VENUE
M3KG-RAG: Multi-hop Multimodal Knowledge Graph-enhanced Retrieval-Augmented Generation
H. Park, J. Seo, J. Mun, H. Park, W. Byeon, S. J. Kim, H. Im, J. Lee, S. Kim
ICLR 2026 TOP VENUE
Judo: A Juxtaposed Domain-oriented Multimodal Reasoner for Industrial Anomaly QA
H. Kang+, W. Lee+, J. Kim, H. Park*
ICML 2025 TOP VENUE
Self-supervised Adversarial Purification for Graph Neural Networks
W. Lee, H. Park*
KDD 2025 TOP VENUE
Harnessing Influence Function in Explaining Graph Neural Networks
H. Jung+, C. Kim+, G. Han, H. Park*
ACL 2025 Findings TOP VENUE
Complex Reasoning in Knowledge Graph Question Answering Through Query Graph Approximation
H. Jeong+, M. Kim+, H. Jung, K. K. Kim*, H. Park*
AAAI 2025 TOP VENUE — Oral (Top 6.1%)
MAMS: Model-Agnostic Module Selection Framework for Video Captioning
S. Lee, I. Y. Chun*, H. Park*
ICLR 2024 TOP VENUE
UNR-Explainer: Counterfactual Explanations for Unsupervised Node Representation Learning Models
H. Kang, G. Han, H. Park*
Neural Networks (IF: 9.657) JOURNAL
Generating Post-hoc Explanations for Skip-gram-based Node Embeddings by Identifying Important Nodes with Bridgeness
H. Park, J. Neville

Interested in our research?

We are always looking for motivated students and researchers to join our lab. Explore open positions or reach out to discuss potential collaborations.

View Open Positions →