Research

In this research area, we focus on developing core technologies for domain-specific AI models and frameworks for medicine. These AI technologies encompass not only task-specific AI models but also foundation models specialized for the medical fields, as well as the development of large multimodal AI models capable of modeling physiology (multimodal physiological AI). Additionally, we design actionable AI models that utilize cutting-edge technologies, such as large multimodal models (LMMs) and Agentic AI. We analyze and validate real-world data (RWD) generated in hospitals from multiple perspectives to effectively model a complex medical environment. 

Current areas of interest: Knowledge editing for medical LLMs, Multimodal RL, Physiology-informed AI agents, Medical world models

In this research area, we focus on studies that uncover knowledge previously unknown in the medical fields, using real-world data (RWD) generated in hospitals. Based on the philosophy of data-driven medicine (also known as data-informed medicine), we aim to question and re-evaluate conventionally accepted protocols, thereby providing objective, data-based, personalized treatment guidelines for individual patients. Additionally, we conduct data-driven research to develop safe medical AI that can assess and mitigate risks, including bias and fairness issues, that may arise during AI model development.

Current areas of interest: Medical knowledge graphs, Medical knowledge verification, Medical AI safety, Fairness/bias in medical AI

In this research field, we investigate efficient verification strategies to ensure the safe deployment of AI models in clinical settings. We analyze the latest requirements mandated by regulatory authorities for medical device licensing and establish corresponding response strategies. Our goal is for the medical AI model we have developed to be used as a safe medical device in clinical settings, thereby directly improving patient outcomes and reducing healthcare costs.