Integrated Ph.D. Student, Korea University
Seokyong Sheem
PhD researcher building multimodal AI systems for digital health, wearable sensing, and medical foundation models.
He studies robust multimodal learning across physiological signals, structured clinical data, language, and behavioral context, with a focus on real-world healthcare and human-centered AI systems.
Research Interests
Multimodal Intelligence
Learning across physiological signals, text, medical imaging, and structured clinical data for robust real-world inference.
Wearable Sensing
Smartwatch-based PPG and IMU sensing pipelines for driver monitoring, behavioral understanding, and health-related prediction.
Human-Centered AI
Designing systems that remain reliable under noisy observations, partial information, and variability in human behavior.
Healthcare AI
Relation-aware learning and foundation modeling for physiological, tabular, and clinically grounded health data.
Selected Projects
Driver State & Takeover Performance
Smartwatch-based multimodal modeling for driver state estimation and takeover performance in conditionally automated driving.
Signals and methods: smartwatch PPG, motion sensing, multimodal fusion, human-centered evaluation.
Medical Foundation Models
Developing relation-aware and topology-aware learning methods for structured health data and medical foundation modeling.
Focus: tabular representation learning, clinical relations, robust transfer across healthcare settings.
Cross-Cohort Tabular Learning
Studying transfer, adaptation, and benchmarking across NHANES, KNHANES, and aging cohorts.
Focus: cross-cohort generalization, external validation, and benchmark design for health AI.
Research Highlights
Smartwatch-Based Driver Monitoring
Developed a CHI 2026 paper on smartwatch-based multimodal AI for predicting driver state and takeover performance.
Medical Foundation Modeling
Building foundation-style models for structured health data with relation-aware and topology-aware learning objectives.
Cross-Cohort Health AI
Benchmarking transfer across NHANES, KNHANES, and aging datasets to improve robustness beyond single-cohort evaluation.
Education
Korea University
Seoul, South Korea
Harbin Institute of Technology
Harbin, China
Selected Publications
MUST: Smartwatch-based Multimodal Framework for Predicting Driver State and Takeover Performance
Introduces smartwatch-based multimodal prediction of driver state and takeover performance in automated driving.
News
Website updated with current projects, research highlights, and publication details.
MUST was accepted to CHI 2026.
Current work focuses on medical foundation models and cross-cohort transfer for structured health data.
Contact
I’m open to research collaboration, internships, and conversations on multimodal AI, digital health, wearable sensing, and medical foundation models.