Edge-AI Solutions For Low-Energy Iot Networks In Smart Cities

Abstract

The exponential growth of the Internet of Things (IoT) systems in smart cities has exacerbated the need of energy-saving, low-latency, and privacy-preserving Artificial Intelligence (AI) applications. In this paper, the authors investigate the performance of three AI deployment paradigms on a simulated low-energy IoT network Cloud-based AI, Edge-AI, and Federated Edge-AI. Four key performance measures were tested with the computation of the energy consumption, latency, bandwidth use, and inference accuracy using quantitative experimentation with fifty IoT nodes within equal computational conditions. The findings have proved that Cloud-based AI is the most accurate (96.9%), but it has high energy and latency overheads since data is processed in a central location. By comparison, Edge-AI consumes 52% less energy and latency with 73% less compared to localized computation efficiency. The Federated Edge-AI paradigm is the best balanced to provide a reduction of 66 percent in energy consumption and 25 percent in bandwidth efficiency, at a small accuracy cost (95.0 percent) relative to the cloud. The other normalized Composite Performance Index merely proves the excellence of Federated Edge-AI (CPI = 0.86) when compared to Edge (0.66) and Cloud (0.25) architectures. The paper summarizes on the fact that federated learning and edge computing is a sustainable, scalable, and privacy-preserving architecture of the next-generation smart city IoT ecosystems. Such hybrid teams are capable of supporting near-cloud intelligence and delivering much higher network resilience and resource efficiency and is a paradigm shift of decentralized AI systems.