Hybrid Cloud-Edge Architectures for LowLatency IoT Machine Learning

The Internet of Things (IoT) is generating massive amounts of data that can be leveraged for machine learning and artificial intelligence applications. However, the centralized cloud computing model faces challenges in meeting the low-latency and real-time requirements of many IoT use cases. This paper proposes and evaluates hybrid cloud-edge architectures that combine edge computing with cloud resources to enable low-latency machine learning for IoT systems. We present a taxonomy of hybrid architectures and develop an analytical model to characterize their latency and resource utilization. Extensive simulations and a real-world case study demonstrate that hybrid approaches can reduce end-to-end latency by up to 75% compared to cloud-only solutions while efficiently utilizing both edge and cloud resources. Key factors impacting performance are analyzed, including data distribution, model complexity, and network conditions. Guidelines for designing effective hybrid architectures for different IoT scenarios are provided. The results show that hybrid cloud-edge computing is a promising paradigm for supporting low-latency IoT machine learning at scale