In this section, we will introduce the stories related to artificial intelligence (AI) with vivid and interesting cartoons. You will learn about the concepts and classifications of machine learning.
I. The concept of machine learning
Human beings need to go through all kinds of learning to understand the world.
When children see a cat for the first time, they may ask their parents what this cute, chubby, bearded animal is. When her parents tell her: This is a cat, she will understand that this creature is a cat. Later, she would recognize all kinds of cats when she met them.
Computer is the same, in the process of learning, can automatically find out the “specific characteristics of the cat”, and form their own set of identification methods, even if there is no problem can be solved according to the original idea.
However, and people’s learning is slightly different, the computer is through a large number of data, find rules, prediction and classification, even if it has not encountered the same type of problems, it can solve!
Machine learning, as the name suggests, machines can learn like children. A machine is no longer simply a tool for running human programs, it can learn by itself! Get smarter!
Machine learning can be roughly divided into three categories: supervised learning, unsupervised learning and reinforcement learning. Let’s take a look at each of these three types of machine learning.
Classification of machine learning
1. Supervised learning
In supervised learning, we feed all the data and the matching answers into a computer, and the computer learns the connections between the characteristic rules of the data and the answers.
In supervised learning:
- Data with answers is necessary because computers are constantly correcting answers and correcting their own problems as they learn (train).
- The amount of data with answers is huge. Computers are not as smart as we think. They need to find their flaws in every mistake. At present, only humans can learn from a small sample or draw inferences from one another.
Supervised learning can be roughly divided into classification problems and regression problems.
(1) Classification problem
In supervised learning, we’ve been using the example of recognizing cats, but it’s actually a classification process, where a computer can classify pictures. Categorization is not limited to images, we can also categorize text content.
The kind of spam that bothers us most is the kind of spam that can be sorted through computer recognition.
Computer spam classification is not as we imagine, directly tell you the answer oh, it will carry out the probability of spam and normal mail labels. For example, this email contains a large number of words “discount”, “promotion”, is spam 92% of the time. We humans make specific distinctions based on probability labels given by computers.
In addition to pictures and words, computers can also distinguish sound.
(2) Regression problem
When it comes to regression, many friends will feel confused. What is regression? Back there? To where?
In fact, regression problems are prediction problems, but in machine learning, they are called regression.
The familiar pokemon – Pokemon attack power can be predicted from historical data.
Regression is the process of finding a trend line that can accurately identify the pair of data from a pile of data to arrive at a specific value.
Specific differences between classification and regression:
We can predict the weather to be sunny, cloudy, rainy and snowy, and this is the classification process. But if you predict a specific weather temperature, it’s regression.
(3) Over-learning and lazy learning (over-fitting and under-fitting problems)
Do we wonder if the more data we give a computer, the better it will be able to classify and regression? The answer is “NO! “”
Computer overlearning is technically called “overfitting”!
I will give a particularly painful example: when I was in junior high school, one day the teacher informed us that we would have a math test for a period of time, so we should review it carefully. I put all the questions after class have done three times, thought, this time my grades will be very good! However, when the exam, paper hair down, found that is a mathematical competition, I have no language coagulate choke…
But, on the other hand, if you can’t be bothered to do the homework, there’s no way to take an exam. This is “lazy learning”, and the result may be not only tears, but also physical abuse. This is called “underfitting”.
Write here, can not help but to our Chinese Confucian culture “” the golden mean” “, secretly admire!
2. Unsupervised learning
Many problems in the world cannot be solved by supervised learning, because many human beings do not know the answers.
Unsupervised learning, as in machine learning, is the process of asking a computer to analyze a pile of unknown data to find structures and rules.
For example, the e-commerce customer classification process is a kind of unsupervised learning. At the beginning, we could not label customers accurately, but gradually, we could distinguish some common features from the purchase records and browsing records of different customer groups and cluster them. The product recommendation service we often receive is the product recommended by e-commerce to a certain type of label users that they may like.
3. Reinforcement learning
Human beings learn from their successes and failures how to achieve their goals smoothly.
I think most of you have had the struggle between playing games and doing homework when your parents were not at home. If you play games, it’s cool now, but if mom and dad suddenly come back… If the teacher checks…. tomorrow Although two sharp swords hung over their heads, many students still picked up the handle and keyboard.
Homework may be a pain now, but if mom and dad come home suddenly, if they do well on a test, it’s a huge benefit.
It must be only after a painful experience, many students think over the pain, or picked up a pen, write homework.
In the same way, a computer can learn by trial and error, experiencing many failures and successes. Failure and success are all about rewards. This is reinforcement learning.
The basic principle of AlphaGo, which makes us familiar with artificial intelligence, is reinforcement learning.
Summary in one chart:
Follow public accounts
【 Pegasus Club 】
▼
Previous welfare concerns about the pegasus public number, reply to the corresponding keywords package download learning materials; Reply “join the group”, join the Pegasus AI, big data, project manager learning group, and grow together with excellent people!
RSVP number “2” machine learning & Data Science must-read classic book with resource pack!
Into AI & ML: Learning machine Learning from Basic Statistics (PDF download)