Federal Grant Award Summary

Rutgers, The State University is conducting foundational theoretical research on memory-constrained machine learning (ML) under a $344,821 Project Grant from the National Science Foundation's (NSF) Division of Computing and Communication Foundations (CFDA 47.070). The award, issued June 1, 2026, with completion targeted for May 31, 2031, addresses critical gaps in understanding how memory limitations affect the data and time requirements for machine learning tasks. The research systematically characterizes the memory requirements and computational capabilities of low-memory algorithms commonly used in practice, including stochastic gradient descent, by drawing on mathematical tools from complexity theory, learning theory, and information theory. The project directly responds to emerging challenges posed by large-scale machine learning systems and large language models (LLMs) where parameter counts increasingly exceed available memory hardware, as well as the growing demand for on-device learning driven by data security and personalization concerns.

The educational component of the award delivers training for undergraduate and graduate students through foundational and advanced courses in theoretical computer science at the intersection of relevant mathematical disciplines, as well as supervised research activities aligned with the project's research themes. These research and educational activities will advance NSF's mission to support investigator-initiated research and education across computing and information science and engineering while building workforce capacity in critical areas of computational theory.