Introduction of this course (Introduction to machine learning)

One 、 Artificial intelligence 、 Machine learning and deep learning

• Artificial intelligence （Artificial Intelligence） $\to$ The goal is
• machine learning （Machine Learning） $\to$ methods
• Deep learning （Deep Learning）$\to$ One of the methods of machine learning

▲ Artificial intelligence 、 The relationship between machine learning and deep learning

Two 、 machine learning （Machine Learning）

So-called Machine Learning The direction of , You just write a program , Then the robot becomes very smart , It can have the ability to learn .

If it is more practical ,Machine Learning The things that were done , You can think that success is looking for a Function, Let the machine have a capability , This ability is based on the information you provide it , It goes to find what we're looking for Function.

$$Machine \ Learning \approx Looking \ for \ a \ Function$$

find function Of framework：

1. Prepare one first function set( aggregate ), This function There are thousands of function, This function set It's called model( Model );
2. Use Training Data Judge this function Is it good or bad? ;
3. An efficient algorithm automatically selects the best function.

▲ Machine Learning Framework The whole process

3、 ... and 、 Machine learning related technologies

3.1 Supervised learning （Supervised Learning）

• Return to （Regression）：Predict continuous valued output
• classification （Classification）：Discrete valued output
• Structured learning （Structured Learning）：lnput and output are both objects with structures

3.2 Semi-supervised learning （Semi-supervised Learning）

Training Data A small amount of Labelled data And a lot of Unlabeled data.
In the technology of semi supervised learning , These are not label Of data, They may also be helpful for learning .

3.3 The migration study （Transfer Learning）

Transfer learning means ： Suppose we have to do the classification of cats and dogs , So do we , Only a few have label Of data. But now we have a lot of data, These are a lot of data There may be label Or maybe not label. But they have nothing special to do with the problem we need to consider now , What we need to distinguish is the difference between a cat and a dog , But there are a lot of pictures of other animals here , You have a lot of irrelevant pictures , What help can they bring . This is the problem of transfer learning .

3.4 Unsupervised learning （Unsupervised Learning）

There is no label, A machine can learn without a teacher .

3.5 Reinforcement learning （Reinforcement Learning）

stay Reinforcement Learning in , We didn't tell the machine what the correct answer was , The machine has only one score , Is it good or bad .

Alpha Go In fact, it is Supervised Learning add Reinforcement Learning To learn . First, use the chess manual to supervise learning , And then do reinforcement learning with another machine .

▲ Machine learning related technologies

Four 、 Why study machine learning

The most important reason is the need to AI Trainer .
In machine learning , You need to pick the right one Model、Loss Function、…, Different Model、Loss Function Suitable for solving different problems , At this time, we need experienced AI Trainer to find the right one Model、Loss Function.