Recently, little Mi found a problem! Why is my style recommended by a song listening software? Why does an orange shopping software recommend many to me, so that I can plant grass directly? Little Mi had an idea. He took out his mobile phone and began to ask Baidu for help. After a search, he found that this is now a very hot AI technology! Not only that, artificial intelligence technology has really existed in our life. For example, the advertising emails we receive every day will be automatically recognized as spam, the photo albums of mobile phones can be grouped by different faces, as well as various voice assistants that are common now. These are related small technologies. Little Mi suddenly became interested and thought, yes! Why don’t you study this thing?! One does not do two endlessly. Since you want to study it, just study it upside down! First of all, Xiao MI is going to start with basic machine learning. If you are also interested, learn with Xiao MI~
1 what is machine learning
What is machine learning? What is machine learning (sorry, what is the stem of happy planet suddenly popped up in little Mi’s mind). So what is machine learning? Little Mi traced the origin and found such a definition given by Tom Mitchell, the godfather of machine learning: for a certain type of task T and performance measurement P, if the performance measured by P on t is self-improvement with experience E, then the computer program can be called learning from experience E. Sure enough, the godfather is the godfather, and his words are different from others. Little MI is dizzy, but it’s not difficult for little Mi! Little Mi looked up all kinds of materials and finally found that it would be more profound to understand this passage through cases! No more nonsense, let’s take an example!
Take e-mail identification as an example. Suppose that the e-mail program will observe whether the received e-mail is marked as spam. In the mail client, click the “spam” button to report that some emails are spam. Based on the messages marked as spam, the program can better learn how to filter spam. In this case, t indicates whether the marked mail is spam, while e is the process of observing whether the marked mail is spam, and P refers to the correct rate of distinguishing spam.
In other words, why can we judge that it will rain later according to the low flying of the dragonfly? Why can we judge the maturity of fruit by the color and shape of its epidermis? That is because we have accumulated a lot of experience, and through the use of experience, we can make effective decisions on new situations. Machine learning is to hand over such a task of learning and prediction to the computer to study various “learning algorithms”. If Xiao MI can’t understand it again, then Xiao MI can only say that Xiao MI can’t do it (the expression bag directly hits you)!
2 supervised learning and unsupervised learning
Let’s continue to absorb the nutrients of new knowledge!
Xiao Mi found that the two main types of learning algorithms are supervised learning and unsupervised learning, which are the most commonly used. Others include reinforcement learning, recommendation system and so on. Since supervised learning and unsupervised learning are two mountains, why not try to cross it!
2.1 supervised learning
Still the same! In primary school, teachers will teach us how to express a formula, what its principle is, and how to use it when doing problems. So what is supervised learning? According to Xiao Mi’s simple understanding, since it is supervised learning, there must be some standards or conditions for “supervision”. Right? Literally, it does mean that. Then little Mi summoned big brother Baidu and found that the example is really God’s assist, which makes me understand it without effort!
Take the national house. Assuming that there is a house in the hands of little Mi that needs to be sold (let little Mi fantasize for a while, hahaha), how much price should be marked on it? The area of the house is 100 square meters, and the price is 1 million, 1.2 million or 1.4 million?
Obviously, little Mi wants to get some relationship between house price and area. Then little MI can investigate some houses with similar house types around and obtain a set of data. Through this set of data, we can draw a straight line or a parabola in line with its price law, so as to determine the house sales price.
This is an example of a supervised learning algorithm. Supervised learning refers to that we give the algorithm a data set, which contains the correct answers, that is, we give the house price a data set. For each sample in this data set, we give the correct price, that is, the actual selling price of the house. The purpose of the algorithm is to give more correct answers. This problem is also called regression problem, because the prediction result is a continuous numerical output, that is, the price in the example is continuous.
Draw inferences from one instance. Since the predicted output can be continuous, it can also be discrete ~ Xiao Mi learned that in view of this kind of situation, it has become a classification problem. Predicting whether Xiao y (hee hee, I hope Xiao y doesn’t mind) is late for work today is nothing more than two results, one is late and the other is not late. Of course, we can continue to classify the problem of being late. One is less than half an hour late, and the other is more than half an hour late. Therefore, such a learning task is a classification problem. Is it easy to understand!
2.2 unsupervised learning
In the above learning about supervised learning, Xiao Mi believes that supervised learning clearly tells what is the “correct answer”, which is the meaning of “supervision”. In unsupervised learning, there are no so-called conditions or standards, so the data sets will be different, with the same label or no label, In other words, unsupervised learning is to let the computer learn by itself (I have to praise that the computer is really a hard-working and smart child!).
Suppose you are given a data set. You don’t know what to do with it, and you don’t know what each data point is. You are just told that there is a data set here. Can you find some structure in it? Anyway, for ordinary humans like little MI, there is really no way to start. But computers are different! For a given data set, the computer determines that the data set in the figure above contains two different clusters (as shown in the figure above) through unsupervised learning algorithm. It can be divided into two different clusters, which is clustering algorithm.
Another example! For example, baidu (hee hee, also mentioned Baidu big brother) search machine learning will pop up various URL links related to it. What Baidu does is to collect all kinds of relevant information on the network every day, and then combine them into a series of special topics.
Of course, clustering algorithm is only one kind of unsupervised learning, and supervised learning is not just as simple as giving two examples. If you want to continue machine learning with Xiaomi, wait for Xiaomi to continue to update~
PS: last but not least! especially! Very! Thank you, Mr. Wu Enda. This machine learning series of little Mi will follow Mr. Wu Enda’s videos one by one. I really have to say that the boss is the boss. Simple examples can be found for complex theories for our understanding and worship!