Mapping Machine Learning to Physics (ML2P) DARPA-PS-25-32
Summary
AI-generated · Sep 28, 2025The program targets making machine learning more power-efficient from the outset, with a focus on edge deployments in resource-constrained battlefield environments. It emphasizes prioritizing energy efficiency as a fundamental design and deployment consideration for ML systems.
It will map ML efficiency directly to physics using precise Joule measurements to enable accurate power and performance predictions across diverse hardware architectures. It also seeks to develop multi-objective optimization functions that balance power consumption with performance metrics and to explore how local optimizations interact through Energy Semantics of ML (ES-ML) to address the energy-aware ML optimization problem.
Machine learning (ML) moves fast, but it needs power. More power than we have, and that s the problem. The Department of Defense faces additional constraints with ML deployments at the edge in resource-limited battlefield environments. The ML2P program is about prioritizing power efficiency consumption right from the start. ML2P will map ML efficiency directly to physics using precise Joule measurements, enabling accurate power and performance predictions across diverse hardware architectures. ML2P will develop multi-objective optimization functions that balance power consumption with performance metrics and discover how local optimizations interact through Energy Semantics of ML (ES-ML) to solve the energy-aware ML optimization problem.
From Solicitation posted on Sep 23, 2025Machine learning (ML) moves fast, but it needs power. More power than we have, and that s the problem. The Department of Defense faces additional constraints with ML deployments at the edge in resource-limited battlefield environments. The ML2P program is about prioritizing power efficiency consumption right from the start. ML2P will map ML efficiency directly to physics using precise Joule measurements, enabling accurate power and performance predictions across diverse hardware architectures. ML2P will develop multi-objective optimization functions that balance power consumption with performance metrics and discover how local optimizations interact through Energy Semantics of ML (ES-ML) to solve the energy-aware ML optimization problem.
From Solicitation posted on Oct 06, 2025Notice history
2-
Solicitation Posted Sep 23, 2025
-
Solicitation LATEST Posted Oct 06, 2025View changes (1)
- Response Deadline: Dec 08, 2025 → Dec 17, 2025
Details
Award Information
Not yet awarded