Energy-Efficient Edge Intelligence: A Comparative Analysis of AIoT Technologies

Modern IoT environments increasingly involve intensive data processing, using advanced algorithms for artificial intelligence, locally on the nodes themselves. That is, for various reasons, it is not possible or desirable to outsource the processing to a remote system - e.g. a local server or remote cloud. For example, in some cases a short response time is necessary. Or, the flow of data to be processed exceeds the capacity of the communication channel. In some cases, the nodes have no reliable or even no connection. The protection of data privacy is becoming another increasingly important reason - for example, in the case of video surveillance of public areas - where any data leaving the device is a potential threat. An additional limitation for IoT devices is the need for energy efficiency, primarily for battery-powered devices. In this paper, we tested the performance, that is, the usability and efficiency, of various IoT devices. Our findings indicate that energy-efficient performance optimization of AI models is possible even in highly constrained scenarios.