table of Contents
Artificial Intelligence and Machine Learning
Machine learning is basically in the same category as the recent hot artificial intelligence. Artificial intelligence is a tall name for machine learning, and machine learning is a technical means to realize artificial intelligence. For example, popular projects in the artificial intelligence era such as face recognition and autonomous driving are all realized by machine learning.
It can be said that mastering the method of machine learning is equivalent to mastering the core connotation of the artificial intelligence era.
In fact, machine learning has been developed for so many years, and it can play a unique role in both theoretical and practical applications. Machine learning has also played an extremely important role in every modern person's life.
For example, when I was learning machine learning, the longest example I cited was spam recognition. Spam recognition is the most classic case in machine learning. By checking the frequency of words in the email, you can initially understand whether an email is spam.
Another example is another case most commonly used in machine learning: the iris data set. By analyzing the length and width of the petals and the length and width of the calyx of a flower, different types of irises can be distinguished.
Machine learning can help people predict the possible category of a case or the possible value of a certain attribute based on some data . Since some attributes are difficult to obtain, this kind of prediction technology is very important.
For example: Is it possible to use some physical examination indicators of a diabetic patient to predict the probability of a diabetic patient suffering from diabetic complications such as blindness, and to prevent such patients from specific diseases in advance?
Types of machine learning
Machine learning has different realization methods according to different purposes, so it is divided into different categories. For beginners, machine learning can be classified in two ways
- According to the data difference: divided into supervised learning and unsupervised learning
- According to the problem to be solved: divided into classification problem and regression problem
Supervised and unsupervised learning
For example, supervised learning is like teaching a child fruit classification, giving the child specific cases one by one, and telling him this is an apple, this is a banana, and this is a pear. When a child encounters a new case, he will recognize whether the new fruit is an apple, a banana or a pear.
The characteristic of supervised learning is to learn and model the labeled data , which is the biggest characteristic of supervised learning which is different from unsupervised learning.
Still in the previous example, if we don't tell the child what kind of fruit each fruit is, can the child distinguish different fruits?
The answer is that it is possible. After all, the difference between fruits is real. If children are smart enough, they can naturally find the difference between fruits and fruits and classify them spontaneously. This is the characteristic of unsupervised learning. The learned data has no labels, so it can only be classified based on the characteristics of the data.
Classification problems and regression problems
Generally speaking, the situation faced by classification problems is that it is hoped to predict the type of an instance through the model obtained by machine learning, while the regression problem is to predict the specific value of an attribute of an instance through the model obtained by machine learning.
For example: predict the color of the clothes Xiaohong wears on Saturday through the content of the dinner that Xiaohong eats on Friday. The color is a categorical variable, so this problem is a classification problem.
Xiaohong's weight is used to predict Xiaohong's weight. The weight here is a numerical variable, so this problem is a regression problem.
What machine learning can do
Machine learning is indeed a very useful tool, but we should know that machine learning is not omnipotent, and even machine learning is not so perfect in the seemingly simple task of distinguishing spam.
Yes, machine learning will make mistakes. This is normal. After all, people also make mistakes. However, through some clever methods, machine learning can reduce the error rate to a very small level. Many scholars can even reduce the error rate of a machine learning model in a specific scenario to less than 10%. There is no free lunch in the world, and it is impossible for a model to hit the world invincible.
In another case, the poor performance of machine learning is not a problem with machine learning. For example, I just gave an example of predicting the color of clothes Xiaohong wears on Saturday. I said that I would use what Xiaohong eats on Friday night to predict the color of clothes she wears on Saturday. From a common sense, there is no absolute connection between what you eat and what you wear. However, in the machine learning method, the two may have a data connection, so machine learning may also be accurately predicted.
This is very metaphysical, and no one can figure it out yet. It is often not easy to find the best model of a certain problem, which involves many aspects. For example, data processing, model selection and parameter selection.
But the situation is not too bad. The next blog will show you the wonderful performance of machine learning models in classic cases. You can try to implement them. I believe this attempt will give you the confidence to explore the deeper use of machine learning.