Mapping Machine Learning to Physics (ML2P)

DARPA's Information Innovation Office (I2O) is soliciting proposals for the Mapping Machine Learning to Physics (ML2P) program, DARPA-PS-25-32. This initiative aims to improve machine learning (ML) model power consumption and performance on existing hardware by integrating energy efficiency throughout the ML lifecycle. The program, with an anticipated budget of up to $5.9M across two phases, seeks to redefine power as a 'first-class citizen' in ML, enabling energy-aware model construction and simulation. Proposers are required to submit abstracts by October 6, 2025, with full proposals due by December 8, 2025. Awards will be Other Transaction (OT) for Prototype Agreements, focusing on developing multi-objective functions for power and performance optimization, and capturing energy semantics of ML. The program emphasizes open-source publication of results and encourages diverse participation, excluding most UARCs and FFRDCs from being prime technical performers unless by exception. Security requirements mandate unclassified submissions, with strict guidelines for handling Controlled Unclassified Information (CUI).

This government agreement, Agreement No. HR0011-XX-9-XXXX, is an Other Transaction for Prototypes between a company and the Defense Advanced Research Projects Agency (DARPA). Its main purpose is to conduct research and development for a prototype, with payments tied to milestone completion. Key articles cover the project's scope, term, management, financial obligations, and dispute resolution. Crucially, the agreement outlines intellectual property rights, including patent and data rights, and imposes restrictions on foreign access to technology developed under the agreement. It also details safeguarding covered defense information, cyber incident reporting, and public release protocols. The document includes attachments for report requirements, milestone schedules, payment instructions via Wide Area Work Flow (WAWF), and definitions. It prohibits the use of certain telecommunications and video surveillance equipment or services from covered foreign countries, emphasizing national security and compliance.

This document outlines a Defense Advanced Research Projects Agency (DARPA) Other Transaction for Prototypes agreement, identified as HR0011-XX-9-XXXX. It details the terms between DARPA and a performer company for a research and development program focused on a specific prototype, requiring a one-third cost share from the performer. The agreement covers the scope, term, project management, administration, obligation, and payment structure, emphasizing milestone-based fixed payments and the use of the Wide Area Workflow (WAWF) system for invoicing. Key provisions address patent and data rights, foreign access to technology, safeguarding covered defense information, cyber incident reporting, and the possibility of follow-on production contracts. It also includes stipulations on property acquisition and disposition, civil rights compliance, security, public release of information, and a prohibition on certain telecommunications and video surveillance equipment from covered foreign countries.

This government agreement, Agreement No. HR0011-XX-9-XXXX, is an "Other Transaction for Prototypes" between DARPA and a Performer for research and development. It outlines the scope, terms, management, and administration of a prototype development program under 10 U.S.C. § 4022. Key aspects include fixed payments based on milestone completion, detailed reporting requirements via the DARPA VAULT website, and specific provisions for intellectual property, including patent rights and data rights with government licenses. The agreement also addresses foreign access to technology, safeguarding covered defense information (CDI) with cyber incident reporting, and a prohibition on certain telecommunications equipment from covered foreign countries. Disputes are resolved through a tiered process, and property disposition is outlined. The Performer is responsible for overall technical and program management, with DARPA providing oversight. The document emphasizes national security interests and compliance with federal regulations.

DARPA-PS-25-32, "Mapping Machine Learning to Physics (ML2P)," is a program solicitation from the Defense Advanced Research Projects Agency (DARPA) aimed at improving machine learning (ML) model power consumption and performance on existing hardware. The program, through its Information Innovation Office (I2O), seeks to develop energy-aware ML by preserving local energy semantics and creating tunable energy-performance objective functions. This amendment extends submission deadlines for abstracts, questions, and proposals, and clarifies the performer budget. The ML2P program is structured into two 12-month phases, with an anticipated total performer budget of $5.9 million. It encourages open-source solutions and seeks to redefine power as a "first-class citizen" throughout the ML lifecycle, moving beyond current disjointed approaches to ML model efficiency.

This government agreement, HR0011-XX-9-XXXX, outlines an "Other Transaction for Prototypes" between DARPA and a performer for research and development. It details the scope, terms, management, and financial obligations, including an expenditure-based payment system with milestones. Key articles cover patent rights, data rights (Government Purpose Rights or Unlimited Rights), foreign access to technology, safeguarding defense information, and cyber incident reporting. The agreement also addresses property disposition and compliance with civil rights and telecommunications prohibitions. All reporting is done via the DARPA VAULT website, emphasizing the protection of sensitive information and the potential for follow-on production contracts.

This document outlines a Defense Advanced Research Projects Agency (DARPA) 'Other Transaction for Prototypes' agreement (HR0011-XX-9-XXXX) for research and development. It details the agreement's scope, term, management structure, and financial obligations, including an expenditure-based payment approach tied to milestones and a 1/3 cost-share with the performer. Key articles cover patent rights, data rights, foreign access to technology, safeguarding covered defense information, cyber incident reporting, and the potential for follow-on production contracts. The agreement emphasizes the performer's responsibility for technical and program management, with DARPA providing oversight and review. It also includes provisions for dispute resolution, property disposition, civil rights compliance, public release of information, and prohibitions on certain telecommunications equipment.

The DARPA-PS-25-32 solicitation outlines the required abstract template for the Machine Learning to Power (ML2P) program, focusing on unclassified submissions for federal grants. Proposers must adhere to strict formatting (12-point Times New Roman, 1-inch margins, 5-page limit excluding cover, table of contents, and bibliography) and content guidelines. Abstracts need to detail the technical approach, including hardware, power measurement infrastructure, objective function finalization, Energy Semantics for Machine Learning (ES-ML) development, and conversion algorithms. Submissions must also cover technical transition plans (emphasizing open-source licensing), team qualifications, and a Rough Order of Magnitude (ROM) cost estimate. All abstracts must be submitted via the DARPA Broad Agency Announcement Tool (BAAT) in PDF or Microsoft Word format, following specific naming conventions and proprietary information markings.