The primary goal of energy forecasting is to improve prediction accuracy, thereby enhancing overall efficiency by cutting waste—a principle that holds equally for on‑device forecasting in mission‑critical edge settings such as military systems. This paper identifies an Accuracy‑Efficiency Paradox: models with higher precision can paradoxically create a net energy deficit. The deficit originates from two sources of consumption—energy spent on edge‑AI inference and the capacity loss caused by battery aging. Battery aging represents a physical dissipation of future usable energy, effectively another form of energy loss. To address this, the authors introduce a Total Cost of Ownership (TCO) framework that treats inference energy and battery degradation as a unified loss metric, aiming to minimize net energy loss. Empirical results show that in thermally sensitive edge environments, the energy saved by the superior precision of complex architectures is often outweighed by the total energy lost due to their high operational intensity.
Blogger's Review: By incorporating battery aging into the energy cost model, the paper offers a more holistic efficiency assessment, providing valuable guidance for the design of edge AI systems.