TinyML

Chapter 5. The “Hello World” of TinyML: Building an Application

A model is just one part of a machine learning application. Alone, it’s just a blob of information; it can’t do much at all. To use our model, we need to wrap it in code that sets up the necessary environment for it to run, provides it with inputs, and uses its outputs to generate behavior. Figure 5-1 shows how the model, on the right hand side, fits into a basic TinyML application.

In this chapter, we will build an embedded application that uses our sine model to create a tiny light show. We’ll set up a continuous loop that feeds an x value into the model, runs inference, and uses the result to switch an LED on and off, or to control an animation if our device has an LCD display.

This application has already been written. It’s a C++ 11 program whose code is designed to show the smallest possible implementation of a full TinyML application, avoiding any complex logic. This simplicity makes it a helpful tool for learning how to use TensorFlow Lite for Microcontrollers, since you can see exactly what code is necessary and very little else. It also makes it a useful template. After reading this chapter, you’ll understand the general structure of a TensorFlow Lite for Microcontrollers program, and you can reuse the same structure in your own projects.

This chapter walks through the application code and explains how it works. The next chapter will provide detailed instructions for building and deploying it to several devices. ...