The artificial intelligence algorithm series teaches artificial intelligence related concepts in a mathematically easy to understand way, which is also the meaning of "for Human" in the English title of this series of books.
Each volume of this series of books can be read independently or as a whole series of books. However, it should be noted that Volume 1 lists the various basic algorithms used in subsequent volumes, and these algorithms themselves are both the basis and practicality.
In January 2020, the first volume of the book series on artificial intelligence algorithms was published.
If you want to build a tall building, you must focus on the foundation. This book will teach artificial intelligence algorithms such as dimension method, distance measurement algorithm, clustering algorithm, error calculation, hill climbing algorithm, linear regression and discrete learning. These algorithms correspond to the processing and identification of specific patterns in the data, and are also the logic behind various recommendation systems in websites such as Amazon and Netflix.
These algorithms are not only the basis of the algorithms introduced in subsequent volumes, they are also very useful in themselves. In this book, the explanations of these algorithms are accompanied by practical numerical calculation examples.
Online learning resources given in volume 1
The first is the Khan Academy, which has collected and sorted out many YouTube videos teaching various mathematical concepts. If you need to review a certain concept, Khan Academy probably has the video explanation you need, and readers can find it by themselves.
Followed by the website "Neural Network Frequently Asked Questions". As a plain text resource, there are a lot of relevant information on neural networks and other artificial intelligence fields:
http://www.faqs.org/faqs/ai-faq/neural-nets/
In addition, the Encog project's wiki page also has a lot of machine learning content, and this content is not limited to the Encog project:
http://www.heatonresearch.com/wiki/Main_Page
Finally, topics related to artificial intelligence and neural networks can also be discussed on Encog's forums. These forums are very active. Your question is likely to get a reply from a community member or even myself:
http://www.heatonresearch.com/forum
In November 2021, "Artificial Intelligence Algorithms (Volume 2): Algorithms Inspired by Nature" was published
The examples in this book use pseudo-code, so everyone can understand them. The GitHub open source library of this book provides versions in several programming languages, so you can strengthen the learning process through practice. You can modify these "no secret" codes yourself. This is not a "black box" demo. If you are a programmer and use one of Java, R, Python, C#, C, Scala and other languages (and possibly more languages), then you can download and run all the sample codes. These codes have been tested and can run normally. You don't need to spend time debugging code, just enjoy the experience of AI learning process.
In this volume, Jeff introduces algorithms such as genetic algorithms, ant colony optimization, and particle swarm optimization, showing their uses (when and why they are useful), and how to implement them. These are important topics. His "Artificial Intelligence Algorithms" series of books introduces some exciting topics that many people would find daunting. Is this book talking about brain surgery? Not! But it involves neural networks and some cutting-edge topics, such as deep belief networks. Please enjoy this book, enjoy this series of books, and enjoy this adventure!
This book introduces algorithms based on genes, birds, ants, cells, and trees. These algorithms can be used to find the best path, identify patterns, find formulas behind data, and even simulate simple lives.
Sometimes, creatures in nature cooperate with each other. If wolves hunt together, birds will migrate in groups. As a programmer, you can design a group of virtual creatures and use them to solve problems together.
At other times, creatures in nature compete with each other. We can use "survival of the fittest" to guide the evolution of the program. Evolutionary algorithms allow multiple potential solutions to compete, multiply and evolve. After many generations, a potentially excellent solution will be evolved.
It is important to remember that we only seek inspiration from nature, not copy nature, and we can deviate from biological processes if necessary. Compared with the processes that can be simulated by advanced computers, actual biological processes are usually much more complicated.
In March 2021, "Artificial Intelligence Algorithm Volume 3 Deep Learning and Neural Networks" was published, and it is currently on the list of new books.
The re-emergence of neural networks is because Hinton G. (2006) proposed a new deep neural network training algorithm. The latest advances in high-speed graphics processing units (GPUs) allow programmers to train neural networks with three or more layers. Programmers gradually realized the benefits of deep neural networks, which prompted the technology to become popular again.
In order to lay the foundation for the rest of this book, we start by analyzing classic neural networks, which are still useful for various tasks. Our analysis includes concepts such as Self-Organizing Map (SOM), Hopfield neural network and Boltzmann machine. We also introduced the FeedForward Neural Network (FFNN) and showed several ways to train it.
A feedforward neural network with many layers becomes a deep neural network. This book contains methods for training deep networks, such as GPU support. We will also explore techniques related to deep learning, such as random dropout, regularization, and convolution. Finally, we demonstrate these technologies through some real examples of deep learning, such as predictive modeling and image recognition.
Neural network route guidance
This book contains various types of neural networks. We will provide these neural networks and their examples to show neural networks in specific problem domains. Not all neural networks are suitable for every problem domain. As a neural network programmer, you need to know which neural network to use for a specific problem.
Here is a high-level guide to the rest of the book, which will guide you through the areas of interest in this book. Table 1 shows the types of neural networks in this book and their applicable problem domains.
The problem domains listed in Table 1 are described as follows.
- Clustering: Unsupervised clustering problem.
- Regression: For regression problems, the neural network must output numbers based on input.
- Classification: For classification problems, the neural network must divide the data points into predefined categories.
- Prediction: Neural networks must predict events in time, such as signals from financial applications.
- Robot: A robot controlled by sensors and motors.
- Vision: Computer Vision (CV) issues require computers to understand images.
- Optimization: An optimization problem that requires a neural network to find the best ranking or set of values to achieve the goal.
The number of checkmarks (√) gives the applicability of each neural network type to the particular problem. If it is not checked, it means that the neural network type cannot be applied to the problem domain.
All neural networks have some common characteristics, such as neurons, weights, activation functions, and layers, which are the building blocks of neural networks. In Chapter 1 of this book, we will introduce these concepts and introduce the basic features common to most neural networks.
I wish you a happy reading!