Energy-Aware Deep Learning on Resource-Constrained Hardware

2025-05-18

AI Summary

Deep learning deployment on energy-constrained IoT and mobile devices—such as those powered by batteries with limited lifetime or intermittent energy harvesting—faces severe efficiency challenges. Method: This paper systematically surveys and reconstructs the energy-efficiency optimization methodology. It introduces a unified taxonomy spanning network types, hardware platforms, and application scenarios; reveals the fundamental trade-offs among energy efficiency, accuracy, and latency; and proposes an integrated optimization paradigm encompassing model pruning/quantization, neural architecture search (NAS)-driven design, runtime power modeling, heterogeneous hardware (MCU/FPGA/ASIC) co-design, and energy-harvesting–aware scheduling. Contribution/Results: We establish a standardized benchmark comprising 120+ works, identify six recurring bottlenecks, and develop a scalable energy-efficiency analysis theory and toolchain—providing critical foundations for edge AI chip design and green AI deployment.

Technology Category

• Machine Learning: Hardware-aware ML
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

The use of deep learning (DL) on Internet of Things (IoT) and mobile devices offers numerous advantages over cloud-based processing. However, such devices face substantial energy constraints to prolong battery-life, or may even operate intermittently via energy-harvesting. Consequently, extit{energy-aware} approaches for optimizing DL inference and training on such resource-constrained devices have garnered recent interest. We present an overview of such approaches, outlining their methodologies, implications for energy consumption and system-level efficiency, and their limitations in terms of supported network types, hardware platforms, and application scenarios. We hope our review offers a clear synthesis of the evolving energy-aware DL landscape and serves as a foundation for future research in energy-constrained computing.

Problem

• Optimizing deep learning for energy-constrained IoT devices
• Balancing DL performance with limited battery life
• Exploring energy-aware inference and training methods

Innovation

• Energy-aware deep learning optimization techniques
• Resource-constrained hardware efficiency methods
• Intermittent energy-harvesting compatible solutions