Artificial Neural Networks with Java

Develop neural network applications using the Java environment. After learning the rules involved in neural network processing, this second edition shows you how to manually process your first neural network example. The book covers the internals of front and back propagation and helps you understand the main principles of neural network processing. You also will learn how to prepare the data to be used in neural network development and you will be able to suggest various techniques of data preparation for many unconventional tasks.

This book discusses the practical aspects of using Java for neural network processing. You will know how to use the Encog Java framework for processing large-scale neural network applications. Also covered is the use of neural networks for approximation of non-continuous functions. In addition to using neural networks for regression, this second edition shows you how to use neural networks for computer vision. It focuses on image recognition such as the classification of handwritten digits, input data preparation and conversion, and building the conversion program. And you will learn about topics related to the classification of handwritten digits such as network architecture, program code, programming logic, and execution.

The step-by-step approach taken in the book includes plenty of examples, diagrams, and screenshots to help you grasp the concepts quickly and easily.

WHAT YOU WILL LEARN

* Use Java for the development of neural network applications
* Prepare data for many different tasks
* Carry out some unusual neural network processing
* Use a neural network to process non-continuous functions
* Develop a program that recognizes handwritten digits

WHO THIS BOOK IS FOR

Intermediate machine learning and deep learning developers who are interested in switching to Java

IGOR LIVSHIN is a senior specialist at Dev Technologies Corp, specializing in developing neural network applications. He worked previously as a senior J2EE architect at two large insurance companies: Continental Insurance and Blue Cross & Blue Shield of Illinois, developing large-scale enterprise applications. Igor published his first book, WebSphere Studio Application Developer 5.0 (Apress), in 2003. He has a master’s degree in computer science from the Institute of Technology in Odessa, Russia/Ukraine.

Part One. Getting Started with Neural Networks

Chapter 1. Learning Neural Network

Chapter Goal: This chapter introduces you with the Artificial Intelligence Neural Networks

Sub-Topics

Biological and artificial neurons

Activation functions

Summary

Chapter 2. Internal Mechanism of Neural Network Processing

Chapter Goal: The chapter explores the inner machinery of neural network processing

Sub-Topics

Function to be approximated

Network architecture

Forward pass calculations

Back-propagation pass calculations

Function derivative and function divergent

Table of most commonly used function derivatives

Summary

Chapter 3. Manual Neural Network Processing

Chapter Goal: Manual neural network processing

Sub-Topics

Example 1. Manual approximation of a function at a single point

Building the neural network

Forward pass calculation

Backward pass calculation

Calculating weight adjustments for the output layer neurons

Calculating weight adjustments for the hidden layer neurons

Updating network biases

Back to the forward pass

Matrix form of network calculation

Digging deeper

Mini-batches and stochastic gradient

Summary

Part Two. Neural Network Java Development Environment

Chapter 4. Configuring Your Development Environment

Chapter Goal: Explain how to download and install a set of tools necessary for building, debugging, testing, and executing neural network applications.

Sub-Topics

Installing Java 8 environment on your Windows machine

Installing NetBeans IDEInstalling Encog Java framework

Installing XChart Package

Summary

Chapter 5. Neural Network Development Using Java Encog

Framework

Chapter Goal: Using Java Encog framework.

Sub-Topics

Example 2. Function approximation using Java environment

Network architecture

Normalizing the input datasets

Building the Java program that normalizes both datasets

Program code

Debugging and executing the program Processing results for the training method

Testing the network

Testing results

Digging deeper

Summary

Chapter 6. Neural Network Prediction Outside of the Training Range

Chapter Goal: Neural network is not a function extrapolation mechanism.

Sub-Topics

Example 3a. Approximating periodic functions outside of the training range

Network architecture for example 3a

Program code for example 3a

Testing the network

Example 3b. Correct way of approximating periodic functions outside of the training range

Preparing the training data

Network architecture for the example 3b

Program code for example 3b

Training results for example 3b

Testing results for example 3b

Summary

Chapter 7. Processing Complex Periodic Functions

Chapter Goal: Approximation of the complex periodic function

Sub-Topics

Example 4. Approximation of a complex periodic function

Data preparation

Reflecting function topology in data

Network architecture

Program code

Testing the network

Digging deeper

Summary

Chapter 8. Approximating Non-Continuous Functions

Chapter Goal: This chapter introduced the micro-batch method that is able to approximate any non-continuous function with high precision results.

Sub-Topics

Example 5. Approximating non-continuous functions

Approximating non-continuous function using conventional network process . . . . . . .

Network architecture

Program code

Code fragments for the training process

Unsatisfactory training results

Approximating the non-continuous function using micro-bach method

Program code for micro-batch processing

Program Code for the getChart() method

Code fragment 1 of the training method

Code fragment 2 of the training method

Training results for micro-batch method

Test processing logic

Testing results for micro-batch method

Digging deeper

Summary

Chapter 9. Approximation Continuous Functions with Complex Topology

Chapter Goal: Neural network has problem approximating continuous functions with complex topology. It is very difficult to obtain a good quality approximation for such functions. This chapter showed that the micro-batch method is able to approximate such functions with high precision results.

Sub-Topics

Example 5a. Approximation of continuous function with complex topology

Network architecture for example 5a

Program code for example 5a

Training processing results for example 5a

Approximation of continuous function with complex topology using micro-batch method

Program code for example 5a using micro-batch method

Example 5b. Approximation of spiral-like functions

Network architecture for example 5b

Program Code for example 5b

Approximation of the same functions using micro-batch method

Summary

Chapter 10. Using Neural Network for Classification of Objects

Chapter Goal: Show how to use neural networks for classification of objects

Sub-Topics

Example 6. Classification of records

Training dataset

Network architecture Testing dataset

Program code for data normalization

Program code for classification

Training results

Testing results

Summary

Chapter 11. Importance of Selecting the Correct Model

Chapter Goal: Explained the importance of selecting a correct working model

Sub-Topics

Example 7. Predicting next month stock market price

Data preparation

Including function topology in the dataset

Building micro-batch files

Network architecture

Program code

Training process

Training results

Testing process

Test processing logic

Testing results

Analyzing testing results Summary

Chapter 12. Approximation of Functions in 3-D Space

Chapter Goal: Using neuron network for approximation of functions in 3-D space.

Sub-Topics

Example 8. Approximation of functions in 3-D space

Data preparation Network architecture

Program code

Processing results

Summary

Part Three. Introduction to Computer Vision

Chapter 13. Image Recognition

Chapter Goal: introduction to the computer vision - the branch of Artificial Intelligence

Sub-Topics

Classification of handwritten digits

Input data preparation

Input data conversion

Building the conversion program

Summary

Chapter 14. Classification of Handwritten Digits

Chapter Goal: Developed a program able to recognize (classify) handwritten digits

Sub-Topics

Network architecture

Program code

Programming logic

Execution

Summary