This is the first day of my participation in Gwen Challenge.
This article is the machine learning course notes of Li Hongyi, course portal 👇. [Machine Learning 2021] Forecast the number of viewers (part 1) – Introduction to basic concepts of machine learning
Tell the machine to find a function
In a nutshell, machine learning is the ability of a machine to find a function. Now that the machine has the ability to find functions, what kind of things can it do? For example, machines can do several things.
Suppose you want to ask a machine to do speech recognition today, and the machine listens to a sound and produces the text corresponding to that sound, then what you need is a function, and the input is the sound signal, and the output is the content of the sound signal. As you can imagine, this function must be very complicated, and obviously it is not an equation that can be written by hand, so we expect to find this function automatically with the help of machine power, which is machine learning.
Image identification
Similarly, a function, which takes in an image, outputs what’s in that image.
To make a machine play go, we can use a function that inputs the position of the white and black spots on the board and outputs the position of the machine’s next move. If it is good enough, we can also make an AlphaGo.
Machine Learning
Regression Regression
If we are looking for a function whose output is a value, that is, a scalar, then the person who finds the function is called Regression.
For example, the input of the function we found may be various indexes related to the prediction of PM2.5, including today’s PM2.5 value, today’s average temperature, today’s average ozone concentration, etc., and its output is tomorrow’s PM2.5 value at noon, which is a Regression task.
Classification of Classification
The task for the machine is multiple choice, and we humans prepare some options, called classes, and the function we are looking for now is to select one of the options that we have set as the output, so the problem, the task is called Classification.
Gmail accounts, for example, have a function that detects when an email is spam. Its input is an E-mail, and its output can be either spam or not spam. The machine has to pick a choice between yes and no, which is called Classification, and of course, Classification doesn’t have to be just two choices, it can have multiple choices.
For example, AlphaGo itself is also a Classification problem, but the Classification options are more. If you think about it, if you have 19 by 19 positions on the board, then the problem actually becomes a multiple choice with 19 by 19 choices. What we want the machine to do is to find a function, and the input of this function is the position of the black and white spots on the checkerboard, and the output is to pick the correct choice from 19 by 19 choices, and this problem is also a classification problem.
Structured Learning
So machines today are not just going to do multiple choice questions, not just output a number, but produce a structured object. For example, Structured Learning, in which a machine draws a picture or writes a text, is called Structured Learning. If we use the anthropomorphic way to say that is to ask the machine to learn to create this thing.