Introduction to Machine Learning with TensorFlow.js

A compact and gentle introduction to Machine Learning in JavaScript. Learn by building 4 applications from scratch, no previous experience required.

👀 Read online   📕 Get the book  

Description

Many exciting things are happening with AI, from which, until recently, JavaScript developers were largely shut out. However, things are changing, now if you can do npm install @tensorflow/tfjs you can do AI.

This absolute beginner course takes someone with no knowledge of Machine Learning and teaches them the basics.

The course will teach how to use the popular TensorFlow.JS library, a complete re-write of the popular TensorFlow package in JavaScript. If you have ever been interested in Machine Learning, if you want to get a taste for what this exciting field has to offer, if you want to be able to talk to other Machine Learning/AI specialists in a language they understand, then this workshop is for you!

You’ll learn:

  • What are Neural Networks and how is it related to Machine Learning?

  • What is TensorFlow.js and how to use it?

  • The essential mathematics.

  • How to build and train a neural network to solve regression and classification tasks.

  • How to find, convert, load and use pre-trained models.

All will be explained in baby steps with a mix of 50% lecture and 50% hands-on lab exercises.

Taught by Asim Hussain, co-organiser of the AI JavaScript London Meetup and co-creator of aijs.rocks, Asim has been teaching JavaScript for many years and is author of the book Angular: From Theory to Practice.

Program outline/syllabus

  1. Introduction

    We will cover an overview of Machine Learning and Neural Networks as well as a history of TensorFlow and TensorFlow.js.

  2. Project 1: Using a pre-trained model

    In this first project you will learn how to use a pre-trained model and build your first AI-powered application.

  3. TensorFlow

    Now we will dig deeper into TensorFlow itself we’ll cover what Tensors are, how to create them with TensorFlow.js and how to perform basic mathematical calculations using Tensors.

  4. Optimization

    In this section we will explain the core function of TensorFlow and what makes up the field of Machine Learning, Optimization. We will learn what a loss function is and how to use TensorFlow.js to optimise some values based on the loss function.

  5. Project 2: Linear & Polynomial regression

    In this lecture we will cover what regression is and when would you use it, why we start with regression and how to build your first regression model.

  6. Project 3: Recognizing handwritten numbers

    In this section we will start using the layers API from TensorFlow.js and build a much more sophisticated application that recognizes handwritten digits. We will then use the same problem and solve it using a variety of different ML algorithms.

  7. Project 4: Transfer Learning

    Finally we bring all that knowledge together into Transfer Learning, we will take a pre-trained model and train a new model on top of it. Transfer Learning is one of the fastest and least computationally intensive ways to make use of Machine Learning in JavaScript.

Curriculum

Introduction Introduction Why do JavaScript developers need to learn TensorFlow.js? Setup Instructions What is a Neural Network? What is TensorFlow? Run Existing Machine Learning Models Pre-Trained Models Image Classification MobileNet Inputs and Outputs Tensors What are Tensors? Creating Tensors Tensor Operations Optimization Regression What is Regression? Linear Regression Calculating loss Linear Regression Application Polynomial Regression Application Neural Networks What is MNIST? Setting up the code Understanding the MNIST training data Creating a densely connected Neural Network Training a Neural Network using the Layers API Using a TensorFlow model in inference mode Convolutional Neural Networks Solving MNIST with Convolutional Neural Networks Improving MNIST Transfer Learning What is Transfer Learning? What is a KNN classifier Transfer Learning with MobileNet and KNN Summary Summary