Edge AI for Energy-Efficient Computing: A Systematic Review of Strategies, Challenges, and Future Directions

Abstract

Background and Purpose: Edge Artificial Intelligence (AI) has emerged as a crucial solution for minimizing power consumption during real-time data processing in computing devices. Given the increasing demand for energy-efficient AI systems, this study systematically reviews Edge AI methods focused on optimizing energy efficiency. The research highlights existing challenges and explores future research directions to enhance Edge AI capabilities. 
Methods: This study employs a systematic review methodology, analyzing recent advancements in Edge AI, including new hardware systems, model optimization techniques, and software development tools. The review considers the interplay between model complexity, hardware constraints, security concerns, and the trade-off between performance and energy efficiency. 
Findings: The analysis reveals that optimizing Edge AI requires a multifaceted approach involving hardware innovations, lightweight models, and adaptive algorithms. Key challenges include balancing computational power with energy constraints, ensuring data security in edge environments, and maintaining real-time processing  capabilities. 
Theoretical Contributions: This research identifies three significant research avenues: (1) integrating neuromorphic computing to enhance efficiency and mimic biological neural processes, (2) leveraging federated learning to improve privacy-preserving model training across distributed edge devices, and (3) developing adaptive AI architectures that dynamically adjust computational resources based on workload demands. 
Conclusions and Policy Implementations: To advance Edge AI, policymakers and industry leaders must prioritize the development of energy-efficient hardware, encourage research into low-power AI models, and establish regulatory frameworks for secure edge computing. Future work should explore interdisciplinary collaborations to further optimize Edge AI performance.