AI FOR HEALTHCARE / MULTIMODAL LEARNING
Pronounced “Hyun-Gyung Lee” · You can call me Yung.
AI Research Scientist · A*STAR BII · Singapore
My research focuses on multimodal learning and reasoning in healthcare, with an interest in population-scale data and longitudinal health trajectories.
I am an AI Research Scientist at A*STAR’s Bioinformatics Institute (BII), based in Singapore. I received my Ph.D. in Artificial Intelligence from KAIST in August 2026, advised by Edward Choi. I am interested in integrating diverse healthcare data and reasoning about health over time, across individuals and populations.
Outside of research, I enjoy traveling, the outdoors, art, and good food.
RESEARCH
Preprint
Improves zero-shot chest X-ray analysis through sentence-centric pretraining that expands clinically meaningful image–text alignments and adapts visual representations to each sentence.
ML4H (Machine Learning for Health)
Connects LLMs with clinically grounded diagnostic tools to reason over chest X-rays in multi-turn conversations, with image-derived measurements and visual evidence that make its reasoning verifiable.
NeurIPS (Conference on Neural Information Processing Systems) · Datasets and Benchmarks Track
Spotlight
NeurIPS 2025 Financial Aid Award
Evaluates how vision–language models reason through a diagnosis—from anatomical grounding and measurements to clinical criteria—using 18,988 QA pairs across 12 chest X-ray tasks.
CHIL (Conference on Health, Inference, and Learning)
Generates view-specific chest X-rays from radiology reports and available X-ray views, combining textual findings with complementary visual evidence to preserve clinically relevant details. Originally known as UniXGen.
BMVC (British Machine Vision Conference)
ICLR Workshop on Deep Generative Models for Highly Structured Data
Learns a shared, quantized representation of images and text to generate semantically consistent image–text pairs without conditioning on either modality.
IEEE JBHI (IEEE Journal of Biomedical and Health Informatics)
Unifies medical image–language understanding and report generation through joint pretraining and a multimodal attention masking scheme, supporting classification, retrieval, visual question answering, and report generation.
* Equal contribution.
BACKGROUND
A*STAR Bioinformatics Institute (BII)
KAIST · Advisor: Edward Choi
KAIST · Advisor: Edward Choi
Kwangwoon University
BEYOND THE PAPERS
CXReasonBench: A Benchmark for Evaluating Structured Diagnostic Reasoning in Chest X-rays
UniXGen: Unified Chest X-ray and Radiology Report Generation Model with Multi-view Chest X-rays
MedViLL: Multi-modal Understanding and Generation for Medical Images and Text via Vision-Language Pre-Training
NumPy and PyTorch introductory hands-on session
Breast cancer classification
COMMUNITY
Guided large vision–language model projects, including building and fine-tuning multimodal and language models.
Supported students in EHR projects on text-based code embeddings for heterogeneous health records and clinical prediction.
Taught programming fundamentals to participants from clinical and biomedical backgrounds.
GET IN TOUCH