Live Stats

Live Stats

Live Stats

Live Stats

The Concert Band is performing during Parents' Weekend in Alden Hall on sunday, September 14th at 1pm.

DATA SCIENCEPhD Dissertation Proposal DefenseRuofan Hu Time: Sep 15, 2025, from 9:00- 10:00 amLocation: Gordon Library Conference Room 303PhD Committee:Prof. Elke A. Rundensteiner, Data Science, WPI. Advisor.Prof. Randy Paffenroth, Mathematical Sciences, WPI.Prof. Fabricio Murai, Data Science, WPI.Prof. Feifan Liu, Population and Quantitative Health Sciences, UMass Chan Medical School. External member. Title: Learning from Weak SupervisionAbstract: Deep learning models often rely on high-quality labeled data, yet such resources are scarce and costly in domains like public health and healthcare. This dissertation addresses the problem of learning from weak supervision, where labels are noisy, incomplete, or coarse. This dissertation focuses on two major regimes: noisy supervision, which involves developing strategies to learn effectively from mixed- and unknown-quality labels, and indirect supervision, which entails designing methods that leverage coarse-grained signals to guide fine-grained tasks, such as rationale extraction and clinical notes summarization. Across four tasks, novel approaches are proposed to reduce dependence on expert annotations, mitigate label noise and imbalance, and exploit existing high-level signals. This work contributes to the development of robust, scalable models that enhance clinical decision support and medical text understanding.