Edge AI in Internet of Things

Abstract

Traditionally, big data, such as social media, online shopping and business informatics are mainly created and stored at mega scale data centers. However, with the proliferation of mobile computing and IoT, the trend is in reversal.

The largely untapped potential of the data trove on the edge will present many novel applications and will drive significant innovation for AI. In addition, by bringing machine learning closer to the sensors, reliance on a network is reduced and much lower latency can be achieved without a round trip to the server. Also, less network utilization would mean lower power consumption, which is advantageous for embedded systems.

In this paper, a literature review on Edge AI is presented. The design and technical details of the resource friendly edge AI model for audio signal processing are presented: the design and configuration of the neural network, as well as flashing and deployment onto the Arduino development board is detailed. Conclusions are drawn by analysis of the results obtained. In addition, proposed ideas for future research are detailed.