Machine Learning for Humans chapter 1: Why Machine Learning Matters

directory

Chapter 1 Why machine learning matters. This chapter Outlines the evolution of AI and machine learning — from the past to the present and into the future.

Chapter II Supervised Learning (I). This chapter introduces linear regression, loss function, overfitting and gradient descent through examples.

Chapter 3 Supervised Learning (II). This chapter introduces two classification methods: logistic regression and SVM.

Chapter four Supervised Learning (III). This chapter introduces non-parametric methods: K nearest neighbor estimation, decision tree, random forest. Knowledge of cross validation, hyperparametric tuning, and integration models.

Chapter 5 Unsupervised learning. This chapter introduces clustering: K-means, hierarchical clustering; Dimension reduction: Principal component analysis (PCA), singular value decomposition (SVD).

Chapter 6 Neural networks and deep learning. This chapter introduces the working principles, application fields and implementation methods of deep learning, and reviews how neural networks draw inspiration from the human brain. In addition, this chapter also involves convolutional neural network (CNN), recursive neural network (DNN) and neural network application cases.

Chapter 7 intensive learning. This chapter introduces the Exploration and Exploitation of reinforcement learning, including Markov decision process, Q-learning, strategy learning and deep reinforcement learning.

Appendix: Best Machine learning Resources. A list of resources for learning machine learning.

preface

Machine Learning for Humans is an e-book widely circulated among Machine Learning enthusiasts abroad. It was first published as a serial article on Medium. Later, due to its superior quality and high reading value, the author suggested to sort the article into an e-book for readers to read for free. Vishal Maini, the author of the book, has a BACHELOR of Arts degree from Yale and is currently working at DeepMind; Co-author Samer Sabri, also a Yale graduate, is currently pursuing a master’s degree in computer science at the University of California, San Diego.

Who should read it?

• Developers who want to quickly keep up with machine learning trends;

• General readers who want to learn the rudiments of machine learning and participate in technology development;

• For all readers interested in machine learning.

The book is open to all free of charge. The book will cover the basics of probability theory, statistics, programming, linear algebra, and calculus, but it will also inspire readers without a background in math.

This book is designed to help readers quickly grasp advanced machine learning concepts in 2-3 hours. If you want to learn more about online courses, important books, related projects, etc., please refer to the suggestions in the appendix.

Chapter 1 Why is machine learning important

More than any other innovation in this century, ARTIFICIAL intelligence will play a bigger role in shaping the future of humanity. Now, if you don’t know about AI, you’re probably falling behind, and in a world full of AI, sometimes we wonder — is this technology? Or magic?

The acceleration of ai is staggering. In the past four decades, it has endured several winters and offered false hope. But this time, with the rapid increase in data storage and computer processing power in recent years, things have changed dramatically.

In 2015, Google trained a conversational intelligence agent (AI) that can not only act as a help desk, providing technical services to customers through natural and authentic verbal communication, but also talk about ethics, express opinions, and answer general fact-based questions.

A Neural Conversational Model

In the same year, DeepMind developed a reinforcement learning agent that could outperform human players in 49 Atari games using pixel images and game scores as input. Soon after, in 2016, DeepMind broke its own record with the release of a new deep reinforcement learning method called A3C.

At the same time, AlphaGo’s defeat of the top go player was another remarkable achievement since a supercomputer first beat a world chess champion two decades ago. Many masters still don’t understand how a machine can master the full details and complexity of a traditional Chinese war-strategy game that has about 10,170 variations, dozens of orders of magnitude more than the total number of atoms in the universe, 1,080.

Defeated Lee Se-dol, South Korea’s top go player

In March 2017, openAI-trained agents developed a new language on their own and achieved their goals more efficiently in collaboration. After that, Facebook’s intelligence even learned to negotiate and lie… Just a few days before writing this book (August 2017), OpenAI’s agents beat the world’s top players in Dota2, a multiplayer online tactical arena game, another milestone.

Dendi (human) aligns with OpenAI (bot)

In fact, AI is also changing our daily lives. Next time you travel to Taiwan, point your phone’s camera at the Chinese menu. Google’s Translate App translates traditional Chinese characters into English in real time.

Google Translate uses convolutional neural networks to Translate text in real time

Ai could also be used in medicine, designing evidence-based treatments for cancer patients, or helping clinicians analyze medical test results, or even directly involved in drug development.

Benevolent’s bold declaration by AI

Machines are taking over some traditionally human roles. The next time you call a hotel service and ask for some toothpaste, don’t be surprised if you answer the door and see a service robot instead of a human.

This book will explore the core machine learning concepts behind these technologies. At the end of this reading, we expect you to be able to describe how these technologies are implemented and to have a rudimentary ability to build similar applications.

Semantic trees: Artificial intelligence and machine learning

One tip: When you come across new knowledge, think of it as a semantic tree — you have to learn the basics, like the trunk and the thick branches, and then drill down into the leaves to see if there’s anything else there. – Elon Musk

Machine learning is one of the many sub-fields of artificial intelligence

Artificial intelligence is the study of intelligent agents that sense their environment, develop strategies and make decisions to achieve their goals. It involves mathematics, logic, philosophy, probability, linguistics, neuroscience, decision theory and many other fields. Many technologies fall under the category of artificial intelligence, such as computer vision, robotics, machine learning and natural language processing.

Machine learning is just one subfield of artificial intelligence. Its goal is for computers to learn on their own. Machine learning algorithms do not require explicit pre-programming rules and models. They identify and observe patterns in data and build models that explain tasks.

The AI effect: What really fits the concept of “artificial intelligence”?

The current standard definition of “artificial intelligence” technology is vague and subject to change. The AI label applies to machines that replace humans in traditional tasks, and interestingly, once computers learn how to work, people tend to deny them intelligence. This phenomenon is known as the “artificial intelligence effect”.

For example, when IBM’s Deep Blue beat Garry Kasparov, the world chess champion, in 1997, it was criticised for using brute force search rather than “real” intelligence. McCorduck once wrote, “Every time someone tries to use a computer to do something new, like playing checkers or solving simple, informal problems, it’s all part of the evolution of the field of artificial intelligence — but critics like to point out, ‘That’s not thinking’…”

Perhaps there is an expression that everyone agrees is the most plausible explanation for artificial intelligence, such as:

“AI is anything unfinished.” Douglas Hofstadter

So, what kind of computer can be called AI? Let’s take an example: Are self-driving cars AI? Put it today, yes; In the future, maybe not. What about a chatbot that can have smooth conversations? Of course it is. It has to be.

Powerful ARTIFICIAL intelligence is about to change the world, and mastering machine learning would be a good place to start

If this was weak ARTIFICIAL intelligence (ANI), which is only suitable for a few specific tasks, we are now making big strides towards general artificial intelligence (AGI), or strong AI. AGI refers to human-level ARTIFICIAL intelligence at all kinds of intellectual tasks. It involves learning, planning, and making decisions in uncertain situations, including communicating in natural language, telling jokes, guiding people, buying and selling stocks… I even reprogrammed myself.

That last one is a big problem. If we build an AI that can improve itself, will it then enter a cycle of constant improvement that will result in an “intelligence explosion” decades later, or someday.

Let’s define a super-intelligent machine as one that is capable of far exceeding the total intellectual activity of any human being. If designing machines is one of these intellectual activities, then superintelligent machines can certainly design better machines; There will no doubt be an “intelligence explosion” and human intelligence will be left far behind. So the first super-intelligent machine is the last thing we need to invent, if the machine is obedient enough to tell us how to control it. — STATISTIcian I.J. Good (source: Wikipedia)

You’ve probably heard of this concept: technological singularities. It emerges from the gravitational singularity near the center of a black hole, a one-dimensional point of infinite density where all the laws of physics that humans understand begin to break down.

We know nothing about what happens inside a black hole, because not even light can escape its gravity. Similarly, we don’t know what will happen in the future after we start the cycle of AI self-improvement

The Future of Humanity Institute recently published a paper on when AI researchers think AGI will emerge, and they found that most researchers believe AI will have a 50% chance of outperforming humans in all tasks within the next 45 years. We’ve spoken privately with some scientists, the more rational ones of whom think AGI is a long way off (the upper limit is “never”), but quite a few think it will grow surprisingly fast — maybe just a few years.

Kurzweil’s 2005 work “Singularity Is Near”. It’s 2017, and posters have to be taken down again

In this context, the emergence of superartificial intelligence (ASI) could be one of the best or worst things that the human species has ever done. It poses the challenge of what AI and humans would want if they were to be friendly.

While these speculations don’t tell us much about the future, one thing is certain: Now is a good time to start understanding how machines think. If we want to go further than wheelchair-bound philosophers thinking about abstract concepts, and be smarter about how we plan and strategy for artificial intelligence, We must understand the way machines see the world — what they want, their underlying biases and definitions of failure, their temperament quirks — in the same detail as psychology and neuroscience, which study how humans learn, make decisions, act and feel.

Over the next few years, we will be focusing on some of the most complex, high-stakes AI issues.

For example, how do we deal with the biases that are further reinforced in existing datasets? The world’s top researchers are still debating the potential risks and benefits of AI. What are we to make of these disagreements? What would happen to human goals in a world without work?

Machine learning is at the heart of our move towards ARTIFICIAL intelligence, and it will transform every industry and have a huge impact on our daily lives. At least conceptually, this is the main reason why machine learning is so important — and why we chose this chapter as the first in the book.

Suggestions for reading this book

You don’t have to read this book from cover to cover. Here are three tips, depending on how much time you have and how much interest you have.

• T-shaped reading. Read each chapter carefully and summarize it at the end of each chapter. The resources provided in the appendix can help readers who have read the book thoroughly to further explore their areas of interest.

• Screen reading. Jump right to the chapter that interests you, and read and study with all your heart.

• 80/20 of the law. Take a look at the book and make notes on key concepts that interest you, and recall them in the dead of night.

Next period: chapter two supervised learning (a).

The original address: www.dropbox.com/s/e38nil1dnl7481q/machine_learning.pdf?dl=0#pageContainer16