Title: Machine Learning Methods for Data Disentanglement and Fusion in Biomedical Applications
Date: August 14, 2025
Time: 1:00 pm – 3:00 pm (EST)
Location: Groseclose 403 and zoom: https://gatech.zoom.us/j/99195465628?pwd=OlEpmY6yP8MLwiKxaWRz9da3c4hP4Q.1
Lingchao Mao
Machine Learning PhD Student
H. Milton Stewart School of Industrial and
Systems Engineering
Georgia Institute of Technology
Committee
1 Dr. Jing Li (ISYE, Georgia Tech) (Advisor)
2 Dr. Jianjun Shi (ISYE, Georgia Tech)
3 Dr. Kamran Paynabar (ISYE, Georgia Tech)
4 Dr. Xiaochen Xian (ISYE, Georgia Tech)
5 Dr. Lauren Steimle (ISYE, Georgia Tech)
6 Dr. Jiajing Huang (Data Science and Analytics, Kennesaw State University)
Abstract
This thesis develops machine learning methods to address challenges in biomedical data analysis, including limited supervision, missing modalities, and high-dimensional temporal dynamics. The proposed models aim to disentangle complex biomedical data and fuse diverse sources of information for more reliable prediction and interpretation. First, a Weakly Supervised Transfer Learning (WS-TL) framework enables personalized tumor cell density prediction from MRI using imprecise labels and domain adaptation. Second, the Multi-Modal Fission Learning (MMFL) model decomposes multi-modal data into globally shared, partially shared, and unique components, with natural extension to incomplete multi-modal settings and showing effectiveness in a case study for Alzheimer’s prediction. Third, DynMoCo applies graph-based and knowledge-informed dynamic community detection to 4D Molecular Dynamics (MD) simulations, uncovering modular, localized, and functionally relevant motions and providing a new lens for interpretation and knowledge-discovery. Together, these contributions advance interpretable and robust learning for biomedical data analysis.
