About

Hi, I'm Yinhua

I'm a postdoctoral fellow at the Innocore AI-CRED Institute, KAIST, advised by Prof. Sungsoo Ahn and Prof. Woo Youn Kim. I did my PhD in Computer Science at Seoul National University with Prof. Sun Kim, where my dissertation on subgraph-informed hierarchical learning received the Outstanding Doctoral Dissertation Award.

I develop graph-based and multi-modal AI for drug discovery and biomedical reasoning, integrating molecular, cellular, sequential, textual, and knowledge-graph signals across the preclinical pipeline — with a growing focus on LLM-based scientific reasoning over biological knowledge.

Research

Research Overview

Multi-modal integration for drug discovery

Coupling molecules with text, gene expression, and knowledge graphs.

View details →

Robust & generalizable graph learning

Reliable under distribution shift and across domains.

View details →
Current

LLM agents for biological reasoning

Reasoning over biology to model how cells respond.

View details →

I integrate the many signals of biology — molecules, cells, text, and knowledge graphs — into AI that generalizes beyond its training data, toward trustworthy, biology-grounded systems for drug discovery and personalized medicine.

Updates

News

  • 2026.05AdaPert was accepted by ICML 2026.
  • 2025.08A paper was accepted by EMNLP Findings 2025.
  • 2025.08Started the AI-CRED postdoctoral fellowship at KAIST, advised by Prof. Sungsoo Ahn.
  • 2025.05A paper was accepted by ICML 2025.
Show more
  • 2025.04Two papers accepted by the MLGenX Workshop @ ICML 2025.
  • 2025.03A paper accepted by IEEE TCBB 2025.
  • 2025.02Won the Outstanding Doctoral Dissertation Award, CSE, SNU.
  • 2024.12Successfully defended my PhD — "Subgraph-informed Hierarchical Learning in Clinical and Biomedical Domains."
  • 2024.09Won the Youlchon AI Star Fellowship 2024.
  • 2024.04One paper accepted by the Computational and Structural Biotechnology Journal 2024.
  • 2024.02One paper accepted by CVPR 2024.
  • 2024.02One paper accepted by the ICLR Tiny Paper track 2024.
  • 2024.02Won the 30th Samsung HumanTech Paper Award.
Publications

Selected Publications

CVPR 2024
CVPR 2024 Improving Out-of-Distribution Generalization in Graphs via Hierarchical Semantic Environments
Yinhua Piao, Sangseon Lee, Yijingxiu Lu, Sun Kim · Project
ACL 2023
ACL 2023 Clinical Note Owns its Hierarchy: Multi-Level Hypergraph Neural Networks for Patient-Level Representation Learning
Nayeon Kim*, Yinhua Piao*, Sun Kim (Long, Oral) · Project
IJMS 2022
IJMS 2022 DRPreter: Interpretable Anticancer Drug Response Prediction Using Knowledge-Guided Graph Neural Networks and Transformer
Jihye Shin*, Yinhua Piao*, Dongmin Bang, Sun Kim, Kyuri Jo · Project
AAAI 2022
AAAI 2022 Sparse Structure Learning via Graph Neural Networks for Inductive Document Classification
Yinhua Piao, Sangseon Lee, Dohoon Lee, Sun Kim · Project
All publications
Recognition

Honors & Awards

  • 2025.02Outstanding Doctoral Dissertation Award, CSE, SNU.
  • 2024.09Youlchon AI Star Fellowship.
  • 2024.0230th Samsung HumanTech Paper Award.
  • 2022.0927th Samsung HumanTech Paper Award (Bronze). [link]
  • 2022.04AIIS Retreat Best Poster Award. [link]
  • 2022.02AAAI Student Scholar Scholarship.
  • 2020–22SNU Global Scholarships II. [link]
Background

Education

  • 2020–24Ph.D. in Computer Science, Seoul National University.
  • 2018–20M.S. in Computer Science, Seoul National University.
  • 2014–18B.S. in Software Engineering, Jilin University.
Research · Theme 1

Multi-modal integration for drug discovery

Most biomedical questions only become answerable when several modalities are read together. I build models that couple molecular structure with text, gene expression, and biological knowledge graphs.

MV-CLAM: Multi-View Molecular Interpretation with Cross-Modal ProjectionEMNLP '25
MethodCross-modal projectionLanguage model
DataMolecular graphSMILESText descriptions
Predicting Drug-likeness via Biomedical Knowledge AlignmentICML '25
MethodKnowledge alignmentOne-class boundary
DataMolecular structureBiomedical KGApproved drugs
Improving Reaction Yield Prediction with Chemical Atom-Level Reaction LearningJCIM '25
MethodAtom-level GNN
DataReactions (atom-mapped)Yield records
DRPreter: Interpretable Anticancer Drug Response PredictionIJMS '22
MethodKnowledge-guided GNNTransformer
DataCell-line gene expressionDrug graphsPathway KG
Research · Theme 2

Robust & generalizable graph learning

A model is only useful if it holds up on data it was never trained on. I design graph methods that stay reliable under distribution shift and transfer across domains.

Improving OOD Generalization in Graphs via Hierarchical Semantic EnvironmentsCVPR '24
MethodInvariant learningHierarchical environments
DataMolecular property (OOD splits)
Clinical Note Owns its Hierarchy: Multi-Level Hypergraph Neural NetworksACL '23 · Oral
MethodMulti-level hypergraph
DataClinical notes (EHR)
Sparse Structure Learning via GNNs for Inductive Document ClassificationAAAI '22
MethodSparse structure learningGNN
DataDocument graphs
Research · Theme 3 · Current

LLM agents for biological reasoning

My current focus: agents that reason over biological knowledge to model cells and predict how they respond to perturbations.

Learning Adaptive Perturbation-Conditioned Contexts for Robust Transcriptional Response PredictionICML '26
MethodPerturbation-conditioned contextRobust prediction
DataPerturbation transcriptomicsGene descriptionsBiological KG
Progressive Multi-Agent Reasoning for Biological Perturbation PredictionPreprint
MethodMulti-agentProgressive reasoningLLM
DataPerturbation tasksBiological knowledge