AI and Machine Learning for Coders

Chapter 12. An Introduction to TensorFlow Lite

In all of the chapters of this book so far, you’ve been exploring how to use TensorFlow to create machine learning models that can provide functionality such as computer vision, natural language processing, and sequence modeling without the explicit programming of rules. Instead, using labeled data, neural networks are able to learn the patterns that distinguish one thing from another, and this can then be extended into solving problems. For the rest of the book we’re going to switch gears and look at how to use these models in common scenarios. The first, most obvious and perhaps most useful topic we’ll cover is how to use models in mobile applications. In this chapter, I’ll go over the underlying technology that makes it possible to do machine learning on mobile (and embedded) devices: TensorFlow Lite. Then, in the next two chapters we’ll explore scenarios of using these models on Android and iOS.

TensorFlow Lite is a suite of tools that complements TensorFlow, achieving two main goals. The first is to make your models mobile-friendly. This often involves reducing their size and complexity, with as little impact as possible on their accuracy, to make them work better in a battery-constrained environment like a mobile device. The second is to provide a runtime for different mobile platforms, including Android, iOS, mobile Linux (for example, Raspberry Pi), and various microcontrollers. Note that you cannot train a model with TensorFlow ...