Federal Grant Award Summary

The National Science Foundation's Division of Computing and Communication Foundations awarded University of Wisconsin - Madison a $352,098 Project Grant (CFDA 47.070: Computer and Information Science and Engineering) effective January 1, 2026 through September 30, 2029 for collaborative research on robustness to distribution shifts in computational imaging. The research develops mathematical frameworks and methods to ensure deep learning models for image reconstruction remain reliable when training and test data conditions differ—a critical gap in current computational imaging applications. The project introduces Robust Score-Based Inversion (ROSI), a foundational approach that leverages score-based generative models to quantify distribution shifts, characterize their effects on reconstruction and sampling performance, and enable principled model adaptation across scientific, engineering, and biomedical imaging applications.

Deliverables include a unified mathematical framework for analyzing score-based model robustness under distribution shifts, open-source code dissemination, curriculum development resources at participating institutions, and knowledge transfer through organized special sessions, workshops, and journal issues for the computational imaging research community. The research directly addresses limitations in current deep learning approaches that assume stable data conditions, providing methodologies applicable across diverse imaging modalities and real-world operational contexts where robust image reconstruction is essential.