Mapping Machine Learning to Physics (ML2P)

The Mapping Machine Learning to Physics (ML2P) program is an initiative of the U.S. Department of Defense, administered by the Defense Advanced Research Projects Agency (DARPA). This solicitation seeks to address the escalating energy demands of machine learning (ML) systems, especially in constrained environments such as battlefield scenarios where power and computational resources are limited. By rethinking how machine learning models are constructed and evaluated, ML2P aims to enable efficient edge deployment by prioritizing energy consumption as a foundational design constraint. The program's core objective is to create a new paradigm where machine learning efficiency is directly aligned with physical principles. It proposes the use of precise Joule-based energy measurements to facilitate accurate predictions of both power usage and performance across various hardware architectures. This shift aims to enable system designers to better anticipate the energy implications of ML workloads and adjust algorithms accordingly. ML2P focuses on developing multi-objective optimization functions that balance power usage with traditional performance metrics. A key innovation within the program is the development of Energy Semantics of Machine Learning (ES-ML), a framework that will uncover how local optimization decisions impact overall energy efficiency and model performance. The initiative will explore complex trade-offs in hardware-software co-design, targeting breakthroughs that improve ML deployment in austere settings. While specific funding amounts are not detailed, the program is structured to support advanced applied research and is open to a wide range of U.S.-based research entities. The solicitation does not stipulate a matching requirement, making it accessible to institutions with limited cost-sharing capacity. Award structure, number of awards, and funding tiers are outlined in supplementary documents accessible via SAM.gov. The application process requires submission by December 8, 2025, at 5:00 PM Eastern Time. Although no pre-application stage is required, applicants are encouraged to follow submission guidelines as outlined in the ML2P solicitation PDF and related templates. DARPA will evaluate proposals based on technical merit, relevance to DoD priorities, and feasibility of execution within the proposed timeline and budget. Applicants may download templates for abstracts and proposals directly from the SAM.gov page. All inquiries should be directed to the program’s official email, ml2p@darpa.mil. The solicitation was published on September 23, 2025, and will remain active until January 7, 2026. As a DARPA initiative, this opportunity is federally funded and represents a strategic effort to advance energy-conscious machine learning systems for national defense applications.

Focus on energy-aware ML design and physical performance metrics; align with DARPA optimization goals.