Machine Learning for Mobile

Practical guide to building intelligent mobile applications powered by machine learning

Revathi Gopalakrishnan, Avinash Venkateswarlu

• Publisher: Packt Publishing
• Year: 2018
• ISBN:
  9781788621427
• Edition: 1
• Topic: Computer Science
• Subtopic: Data Modelling & Design
• Index: Computer Science

Leverage the power of machine learning on mobiles and build intelligent mobile applications with ease

Key Features

• Build smart mobile applications for Android and iOS devices
• Use popular machine learning toolkits such as Core ML and TensorFlow Lite
• Explore cloud services for machine learning that can be used in mobile apps

Book Description

Machine learning presents an entirely unique opportunity in software development. It allows smartphones to produce an enormous amount of useful data that can be mined, analyzed, and used to make predictions. This book will help you master machine learning for mobile devices with easy-to-follow, practical examples.

You will begin with an introduction to machine learning on mobiles and grasp the fundamentals so you become well-acquainted with the subject. You will master supervised and unsupervised learning algorithms, and then learn how to build a machine learning model using mobile-based libraries such as Core ML, TensorFlow Lite, ML Kit, and Fritz on Android and iOS platforms. In doing so, you will also tackle some common and not-so-common machine learning problems with regard to Computer Vision and other real-world domains.

By the end of this book, you will have explored machine learning in depth and implemented on-device machine learning with ease, thereby gaining a thorough understanding of how to run, create, and build real-time machine-learning applications on your mobile devices.

What you will learn

• Build intelligent machine learning models that run on Android and iOS
• Use machine learning toolkits such as Core ML, TensorFlow Lite, and more
• Learn how to use Google Mobile Vision in your mobile apps
• Build a spam message detection system using Linear SVM
• Using Core ML to implement a regression model for iOS devices
• Build image classification systems using TensorFlow Lite and Core ML

Who this book is for

If you are a mobile app developer or a machine learning enthusiast keen to use machine learning to build smart mobile applications, this book is for you. Some experience with mobile application development is all you need to get started with this book. Prior experience with machine learning will be an added bonus

Fritz

We have gone through mobile machine learning SDKs offered by Google—TensorFlow for mobile—and Apple—Core ML—in the previous chapters and got a good understanding of them. We looked at the basic architecture of those products, the key features they offer, and also tried a few tasks/programs using those SDKs. Based on what we have explored on the mobile machine learning frameworks and tools so far, we will be able to identify a few gaps that make it difficult to carry out mobile machine learning deployments and subsequent maintenance and support of those deployments. Let me list a few for you:

• Once we create the machine learning model and import it into the Android or iOS application, if there is any change that needs to be done to the model that was imported into the mobile application, how do you think this change will be implemented and upgraded to the application that is deployed and being used in the field? How is it possible to update/upgrade the model without redeploying the application in mobile application stores—the App Store or Play Store?
• Once the machine learning model is in the field and is being used by users in the field, how do we monitor the performance and usage of the model in real-time user scenarios?
• Also, you might have experienced that the process and mechanism to use the machine learning models in iOS and Android is not the same. Also, the mechanism to make the machine learning models created using a variety of machine learning frameworks, such as TensorFlow, and scikit-learn and, in order to make it compatible with TensorFlow Lite and Core ML is different. There is no common process and usage pattern that developers can follow to create and use these models across frameworks. We feel that if there was a common approach to use these machine learning models from different vendors using the same process and mechanism, it would be a lot more simple.

An attempt has been made by the Fritz platform to answer all the previously mentioned gaps observed in machine learning model usage and deployment. Fritz, as a machine learning platform, tries to provide solutions to facilitate machine learning model usage and deployment for mobile applications. It is a mobile machine learning platform with ready-to-use machine learning features, along with options to import and use custom ML models—TensorFlow for mobile and Core ML models.

So, in this chapter, we will be going through the following in detail:

• Understanding the Fritz mobile machine learning platform, its features, and its advantages.
• Exploring Fritz and implementing an iOS mobile application by using the regression model we already created using Core ML.
• Exploring Fritz and implementing an Android mobile application by using the sample Android model we created in Chapter 3, Random Forest on iOS, using TensorFlow for mobile.

Introduction to Fritz

Fritz is a free end-to-end platform that enables us to create machine learning-powered mobile applications easily. It is a platform that enables on-device machine learning, that is, it helps to create mobile machine learning applications that can completely work on mobile devices. It supports both iOS and Android platforms.

Prebuilt ML models

Fritz provides built-in ML models that can be directly used in mobile applications. Here are the two important models that Fritz supports:

• Object detection: You can identify objects of interest in an image or each frame of a live video. This helps you to know what objects are in an image, and where they are within the image. The object-detection feature makes predictions completely on-device and requires no internet connection.
• Image labeling: You can identify the contents of an image or each frame of live video. This also works completely offline and requires no internet connection.

Ability to use custom models

Fritz provides us with the ability to import models built for Core ML, TensorFlow for mobile, and TensorFlow Lite into mobile applications and provides APIs that can interact with these models directly.

Model man...

Table of contents

• Title Page
• Copyright and Credits
• About Packt
• Contributors
• Preface
• Introduction to Machine Learning on Mobile
• Supervised and Unsupervised Learning Algorithms
• Random Forest on iOS
• TensorFlow Mobile in Android
• Regression Using Core ML in iOS
• The ML Kit SDK
• Spam Message Detection
• Fritz
• Neural Networks on Mobile
• Mobile Application Using Google Vision
• The Future of ML on Mobile Applications
• Question and Answers
• Other Books You May Enjoy