A Methodology to Assess Power Modeling in Energy-Aware Federated Learning on Heterogeneous Mobile Devices
arXiv:2605.27601v1 Announce Type: cross Abstract: Estimating CPU power on heterogeneous ARM-based commodity devices is challenging due to limited access to CPU's voltage domains. As a result, state-of-the-art energy-aware Federated Learning (FL) frameworks typically rely on simplified approximate power models to estimate computation energy, rather than the more accurate analytical CMOS-based model. To bridge this gap, we propose a reproducible CPU power estimation methodology combined with a rail-to-cluster mapping technique to retrieve cluster-level supply voltage. We evaluate our approach on
The proliferation of distributed AI and federated learning on edge devices necessitates more accurate power modeling for efficient and sustainable AI compute.
Improved power estimation in federated learning directly impacts the efficiency, reliability, and deployment costs of AI inference on a vast array of mobile and edge hardware.
The ability to accurately model power consumption at a granular level on heterogeneous devices allows for more optimized energy-aware AI systems, extending battery life and reducing operational costs.
- · Edge AI developers
- · Mobile device manufacturers
- · Federated Learning platforms
- · Energy-efficient AI hardware
- · Inefficient power modeling techniques
- · High-energy AI models on edge
- · Hardware not optimized for power awareness
More accurate energy consumption predictions for AI models running on billions of edge devices allow for better resource allocation and longer device lifetimes.
This leads to a greater adoption of on-device AI for privacy-sensitive applications, reducing reliance on cloud infrastructure for certain tasks.
The overall energy footprint of AI, especially at the edge, could significantly decrease, contributing to broader sustainability goals for the computing industry.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG