Igor Livshin - Artificial Neural Networks with Java: Tools for Building Neural Network Applications

This Apress imprint is published by the registered company APress Media, LLC part of Springer Nature.

The registered company address is: 1 New York Plaza, New York, NY 10004, U.S.A.

Artificial intelligence is a rapidly advancing area of computer science. Since the invention of computers, we have observed intriguing phenomena. Tasks that are difficult for humans (such as heavy computations, searching, memorizing large volume of data, and so on) are easily done by computers, while tasks that humans are naturally able to do and do quickly (such as recognizing partially covered objects, intelligence, reasoning, creativity, invention, understanding speech, scientific research, and so on) are difficult for computers. It seems that each of us has a super computer in our head.

Artificial intelligence as a discipline was born in the 1950s. However, the neural network architecture known at that time used perceptions with a linear activation function, so it was unable to solve nonlinear problems. This reason and the lack of computing power made AIs early start unsuccessful. AI was revived in 19741980 but eventually failed again.

A new nonlinear network architecture developed after that second failure and a tremendous increase in the machines computing power finally contributed to the phenomenal success of AI in 1990s. Gradually, AI became capable of solving many industrial-grade problems such as image recognition, speech recognition, natural language processing, pattern recognition, prediction, classification, self-driving cars, robotic automation, and so on.

Tremendous success of AI recently triggered all types of unwarranted speculations. You can read some discussions about robots of the near future matching and exceeding the intelligence of humans. We need to remember that currently AI is a set of clever mathematical and processing methods that let computers learn from the data they process and apply this knowledge to solving many important tasks. A lot of things that belong to humans such as intelligence, emotion, creativity, feeling, reasoning, and so on, are still outside of AI knowledge.

However, things are rapidly changing. In recent years, computers have become so good at playing chess that they reliably beat their human counterparts. That is not surprising, because their creators taught the program centuries of accumulated human experience in chess. Now, the world computer chess championship is established where the machines compete against each other. One of the best chess-playing programs called Stockfish 8 won the world computer chess championship in 2016.

Several years ago Google developed a chess playing program called AlphaZero, which defeated the Stockfish 8 program in the 2017 world computer chess championship. The amazing part of this is that no one taught AlphaZero the chess strategies, as had been done previously during development of other chess-playing programs. Instead, it used the latest machine learning principles to teach itself chess by playing against itself. It took the program four hours of learning chess strategies (while playing against itself) to beat Stockfish 8. Self-teaching is the new milestone achievement of artificial intelligence.

AI has many branches. This book is dedicated to one of them: neural networks. Neural networks enable computers to learn from observational data and make predictions based on that knowledge. This book is about neural networks training and using the training for function approximation, prediction, and image classification.

I would like to thank Celestin Suresh John, Apress executive editor, for helping me make this project a reality. Lots of thanks to Aditee Mirashi, Apress associate editor, and Sourav Bhattacharjee and Aakash Kag, technical reviewers. All contributed greatly to the technical accuracy and style of the book.