What this book covers

Chapter 1, Getting Started with Deep Learning, explains some basic concepts of machine learning and artificial neural networks as the core of deep learning. It then briefly discusses existing and emerging neural network architectures. Next, it covers various features of deep learning frameworks and libraries. Then it shows how to solve Titanic survival prediction using a Spark-based Multilayer Perceptron (MLP). Finally, it discusses some frequent questions related to this projects and general DL area.

Chapter 2, Cancer Types Prediction Using Recurrent Type Networks, demonstrates how to develop a DL application for cancer type classification from a very-high-dimensional gene expression dataset. First, it performs necessary feature engineering such that the dataset can feed into a Long Short-Term Memory (LSTM) network. Finally, it discusses some frequent questions related to this project and DL4J hyperparameters/nets tuning.

Chapter 3, Multi-Label Image Classification Using Convolutional Neural Networks, demonstrates how to develop an end-to-end project for handling the multi-label image classification problem using CNN on top of the DL4J framework on real Yelp image datasets. It discusses how to tune hyperparameters for better classification results.

Chapter 4, Sentiment Analysis Using Word2Vec and the LSTM Network, shows how to develop a hands-on deep learning project that classifies review texts as either positive or negative sentiments. A large-scale movie review dataset will be used to train the LSTM model, and Word2Vec will be used as the neural embedding. Finally, it shows sample predictions for other review datasets.

Chapter 5, Transfer Learning for Image Classification, demonstrates how to develop an end-to-end project to solve dog versus cat image classification using a pre-trained VGG-16 model. We wrap up everything in a Java JFrame and JPanel application to make the overall pipeline understandable for making sample object detection.

Chapter 6, Real-Time Object Detection Using YOLO, JavaCV, and DL4J, shows how to develop an end-to-end project that will detect objects from video frames when the video clips play continuously. The pre-trained YOLO v2 model will be used as transfer learning and JavaCV API for video frame handling on top of DL4J.

Chapter 7, Stock Price Prediction Using the LSTM Network, demonstrates how to develop a real-life plain stock open, close, low, high, or volume price prediction using LSTM on top of the DL4J framework. Time series generated from a real-life stock dataset will be used to train the LSTM model, which will be used to predict the price only 1 day ahead at a time step.

Chapter 8, Distributed Deep Learning on Cloud – Video Classification Using Convolutional LSTM Network, shows how to develop an end-to-end project that accurately classifies a large collection of video clips (for example, UCF101) using a combined CNN and LSTM network on top of DL4J. The training is carried out on Amazon EC2 GPU compute cluster. Eventually, this end-to-end project can be treated as a primer for human activity recognition from video or so.

Chapter 9, Playing GridWorld Game Using Deep Reinforcement Learning, is all about designing a machine learning system driven by criticisms and rewards. It then shows how to develop a GridWorld game using DL4J, RL4J, and neural QLearning that acts as the Q function.

Chapter 10, Developing Movie Recommendation Systems Using Factorization Machines, is about developing a sample project using factorization machines to predict both the rating and ranking of movies. It then discusses some theoretical background of recommendation systems using matrix factorization and collaborative filtering, before ping the project implementation using RankSys-library-based FMs.

Chapter 11, Discussion, Current Trends, and Outlook, wraps up everything by discussing the completed projects and some abstract takeaways. Then it provides some improvement suggestions. Additionally, it covers some extension guidelines for other real-life deep learning projects.