Tiny Machine Learning Basic Concepts

Abstract

The artificial intelligence field is booming in unexpected ways and has entered into diverse research areas, unlike conventional machine learning methods, which often require substantial computational power, high-performance servers, or network nodes. Tiny machine learning (TinyML) techniques execute efficiently on small embedded devices such as sensors and microcontrollers. Machine learning capabilities are pushed to the extreme edge, enabling fast real-time responsiveness. TinyML is a composition of mainly software, hardware, and algorithms. The hardware comprises Internet of Things (IoT) devices with or without hardware accelerators. The current scenario advocates the use of TinyML as a possible solution to reduce carbon footprints and considers it extremely useful in low-profile devices with limited computing power. TinyML represents a significant paradigm shift by making it possible to execute ML tasks directly on IoT devices with strict limitations. In this chapter, basic concepts of TinyML, components, applications, and benefits are discussed. This chapter also discusses the challenges and issues in implementing systems using TinyML.