[Abstract]

Artificial intelligence has shown great promise in medical image and healthcare data analysis. However, real-world medical datasets often suffer from limited annotations, heterogeneous data distributions across institutions, and strict privacy constraints that prevent centralized data sharing. These challenges make it difficult to directly apply conventional supervised learning methods that rely on large-scale, fully labeled datasets. In this talk, I will introduce recent approaches for developing robust AI systems that can effectively learn from weakly labeled and heterogeneous medical data. The presentation will cover weakly supervised learning methods for leveraging imperfect clinical annotations, federated learning frameworks that enable collaborative model training without sharing sensitive patient data, and emerging medical foundation models that learn generalizable representations from large-scale medical datasets. I will also discuss recent advances in instruction-free tuning strategies that allow foundation models to adapt to new medical tasks with minimal supervision. Together, these approaches aim to enable scalable and trustworthy AI solutions for real-world healthcare environments.

[Biography]

Prof. Park is an Associate Professor in the Department of Computer Science and Engineering at POSTECH. Before joining POSTECH, he was a faculty member at DGIST for nine years (Feb. 2017–Feb. 2026). He previously completed postdoctoral training at SRI International with Prof. Kilian Pohl and at the Biomedical Research Imaging Center at the University of North Carolina with Prof. Dinggang Shen. Prof. Park received his Ph.D. from Seoul National University in 2014 and his B.S. from Yonsei University in 2008.