“Resource-Aware Deep Learning: Neural Network Optimization for Edge Devices: A Review”

Abstract

The rapid growth of deep neural networks (DNNs) has led to remarkable improvements in accuracy and scalability but at the expense of high energy consumption, making them difficult to deploy on resource-constrained edge devices. With the increasing demand for real-time and privacy-preserving AI applications in healthcare, autonomous systems, and smart cities, energy-efficient deep learning has become a critical research frontier. This paper reviews the historical progression and state-of-the-art strategies for optimizing neural networks to run effectively on edge hardware. Key approaches include model compression, pruning, and quantization, which significantly reduce storage and computational costs while maintaining accuracy. Lightweight architectures such as MobileNet, ShuffleNet, and EfficientNet have further enhanced the feasibility of on-device inference. Additionally, hardware–software co-design, federated edge learning, and neuromorphic computing provide promising pathways toward ultra-low-power AI systems. Despite these advances, challenges remain in balancing accuracy-efficiency trade-offs, addressing hardware heterogeneity, and ensuring robustness against adversarial attacks. This paper highlights current methodologies, identifies key challenges, and outlines future directions, including sustainable AI metrics and adaptive neural models. By bridging algorithmic innovation with energy-aware design, the study emphasizes the path toward scalable, sustainable, and real-world deployment of deep learning on edge devices.