The Mapping Machine Learning to Physics (ML2P) program aims to prioritize power efficiency in machine learning (ML) by mapping ML efficiency directly to physics using precise Joule measurements. This will enable accurate power and performance predictions across diverse hardware architectures. The program will develop multi-objective optimization functions that balance power consumption with performance metrics and discover how local optimizations interact to solve the energy-aware ML optimization problem. Key aspects of the program include: * Developing energy-aware ML models that preserve local energy semantics * Creating tunable energy-performance objective functions * Enabling accurate predictions of power and performance for future ML models * Driving the development of more energy-efficient compute hardware for the Department of Defense The program will explore two prerequisite areas: developing multi-objective functions for power and performance optimization, and discovering the power-performance interactions between local optimizations via capturing the Energy Semantics of ML (ES-ML).