 |
|
|
Clockwise from top left: ECE faculy membes Cunxi Yu, Ang Li, Reza Ghanadan, Ankur Srivastava |
|
A University of Maryland Department of Electrical and Computer Engineering (ECE) team has officially launched a new $1.8 million research project under the Defense Advanced Research Projects Agency’s (DARPA) Mapping Machine Learning to Physics (ML2P) program, which kicked off June 5, 2026. The project sets out to make energy consumption a first-class design objective in modern artificial intelligence (AI) systems.
The project is led by Principal Investigator Cunxi Yu, with co-Principal Investigators Ang Li, Reza Ghanadan, and Ankur Srivastava, all faculty members in UMD ECE, and is carried out in collaboration with Northrop Grumman Corporation.
As AI models scale, their energy demands have become a defining constraint, especially for systems that must operate on limited power at the edge. DARPA’s ML2P program responds by reframing power consumption as a first-class consideration throughout the machine-learning lifecycle, mapping model performance to physical electrical characteristics measured in joules so that AI can strike the right balance between accuracy and energy use.
“Energy has long been an afterthought in machine learning. Our goal is to make it something we can measure precisely, reason about formally, and optimize directly, across any hardware,” said Yu.
Ultimately, the project positions UMD ECE at the forefront of a new direction for AI, one where energy is treated as a measurable, first-class design goal rather than an afterthought. By the close, the team aims to deliver both new scientific understanding and practical, openly available tools that help make AI more efficient and deployable wherever power is limited, from the data center to the edge.
August 13, 2026
|
