16 Deep Learning

Objectives

In this chapter you’ll:

• Understand what a neural network is and how it enables deep learning.

• Create Keras neural networks.

• Understand Keras layers, activation functions, loss functions and optimizers.

• Use a Keras convolutional neural network (CNN) trained on the MNIST dataset to recognize handwritten digits.

• Use a Keras recurrent neural network (RNN) trained on the IMDb dataset to perform binary classification of positive and negative movie reviews.

• Use TensorBoard to visualize the progress of training deep-learning networks.

• Learn what reinforcement learning, Q-learning and OpenAI Gym are and investigate them in exercises.

• Learn which pretrained neural networks come with Keras.

• Understand the value of using models pretrained on the massive ImageNet dataset for computer vision apps.

Outline

1. 16.1 Introduction

   1. 16.1.1 Deep Learning Applications

   2. 16.1.2 Deep Learning Demos

   3. 16.1.3 Keras Resources

2. 16.2 Keras Built-In Datasets

3. 16.3 Custom Anaconda Environments

4. 16.4 Neural Networks

5. 16.5 Tensors

6. 16.6 Convolutional Neural Networks for Vision; Multi-Classification with the MNIST Dataset

   1. 16.6.1 Loading the MNIST Dataset

   2. 16.6.2 Data Exploration

   3. 16.6.3 Data Preparation

   4. 16.6.4 Creating the Neural Network

   5. 16.6.5 Training and Evaluating the Model

   6. 16.6.6 Saving and Loading a Model

7. 16.7 Visualizing Neural Network Training with TensorBoard

8. 16.8 ConvnetJS: Browser-Based Deep-Learning Training and Visualization

9. 16.9 Recurrent Neural Networks for Sequences; Sentiment Analysis with the IMDb Dataset

   1. 16.9.1 Loading the IMDb Movie Reviews Dataset

   2. 16.9.2 Data Exploration

   3. 16.9.3 Data Preparation

   4. 16.9.4 Creating the Neural Network

   5. 16.9.5 Training and Evaluating the Model

   6. 16.10 Tuning Deep Learning Models

   7. 16.11 Convnet Models Pretrained on ImageNet

   8. 16.12 Reinforcement Learning

   9. 16.12.1 Deep Q-Learning

   10. 16.12.2 OpenAI Gym

   11. 16.13 Wrap-Up

1. Exercises