Federal Grant Award Summary

The University of Michigan received a $324,789 Project Grant awarded October 1, 2025, under the Computer and Information Science and Engineering program (CFDA 47.070) from the National Science Foundation's Division of Computing and Communication Foundations to develop foundational theories and algorithms enabling resource-efficient machine learning (ML). The project delivers theoretically-grounded methodologies that optimize ML model design under application-specific resource constraints, addressing the computational and data limitations that restrict ML deployment across healthcare, mobile computing, and other domains. Primary deliverables include improved loss functions and regularization techniques that clarify tradeoffs between data volume, label quality, and model accuracy; model compression schemes that balance model size with performance; and computationally-efficient algorithms for identifying optimal model architectures. These outcomes enable ML systems to operate with reduced human-annotated data requirements, lower computing power demands, and broader hardware compatibility.

The research incorporates an applied demonstration component through the development of efficient hydrological models for water resource prediction and management, validating the practical utility of the resource-efficient algorithms. The award includes a substantial education component featuring mentoring of undergraduate students, development of new undergraduate and graduate courses, and live-broadcast lectures via publicly accessible online platforms to disseminate research findings and build workforce capacity in efficient machine learning methodologies. The project completion date is January 31, 2026.