Machine learning foundation plan 0-2: what is machine learning? What does it have to do with AI?

0 Write it at the front

The machine learning strong foundation program focuses on depth and breadth , Deepen the understanding and application of machine learning models .“ deep ” The mathematical principle behind the detailed derivation algorithm model ;“ wide ” In analyzing multiple machine learning models ： Decision tree 、 Support vector machine 、 Bayesian and Markov decision 、 Strengthen learning, etc .

details ： Machine learning strong foundation program

1 What is machine learning ？

The first question is ： What is such a big concept of machine learning ？ What is the difference with traditional programming logic ？

1.1 Definition

In a word ： Machine learning is a means of studying how to calculate , The discipline of using experience generation models to improve the performance of the system itself

There are several abstract concepts , What is experience ？ What is a model ？ How to improve the performance of the system ？ Next, through an example to illustrate concretely .

In Economics , There is a close relationship between personal income and consumption . More income , The higher the consumption level ; Less income , The consumption level is also low . From the perspective of a society as a whole , Average personal income x And average consumption y The relationship between them is roughly linear . Now let's look at the relationship between passerby A's income and consumption level

income x	Consumption level y
1.0	0.5
1.2	0.8
1.5	1.3

The problem now is ： In a society , The income level is 1.5 The crowd , How about the consumption level ？

First , Visualize the data of passerby a in the coordinate system

In middle school , We learned a concept , It's called Linear least squares , Its realization idea is ： First, give a point set , Then use a straight line to fit this point set , Minimize the distance from all points to this line —— Minimum error . Now let's make such a straight line .

What can this straight line do ？ Can be used to predict ！ We can now judge under any income level according to this straight line , How about your consumption level .

In this case , These three points are experience , We predict it based on the experience of passerby a , In machine learning , Experience shows data ; And this line is called the model , The model reflects Potential patterns and laws contained in data .

Someone will say , This is bullshit , With just three data of one person, we can do such a big task as income level prediction ？

That's true , So the amount of data needed for machine learning is very large , Only with the support of big data , The obtained model has practical significance . for instance , We also sampled the data of some people , The result is as follows

Apply the least square method again , We found that the straight line changed . More data 、 The higher the quality , The stronger its practical significance 、 The less chance . So the machine learning model will become effective ( It is called generalization , But this concept will be discussed later ), We can think that the system performance has been improved .

This is in line with our human cognitive law ！ As we go through more and more , Experience means more and more experience , The understanding of things is also more and more profound . Maybe in the past, we had a wrong understanding of something , It will be corrected in the next stage of life , This is the truth of machine learning .

in fact , We have been exposed to machine learning since high school , The least square method mentioned above is the theoretical basis of linear regression and other models introduced later ！

So the concept of machine learning is not as big as you think , Namely Machines imitate human learning , The process of obtaining potential objective laws from data .

1.2 Programming logic

Previously, we talked about the concept of data and experience , This actually belongs to Statistical significance .

What does that mean ？ The data in the world cannot be exhausted , There are always oneortwo cases , But as long as we are in a statistical sense , It is enough to have enough probability to show that the model conforms to cognition .

So the idea of probability and statistics will run through machine learning . Traditionally, if we input deterministic instructions to make the computer work , This is a Causal , But in machine learning , What we input is test data , The result is the prediction of the maximum probability , For example, the above example of income consumption , Assume income 1.5, The model tells me consumption 1.2, Is everyone 1.2 Consumption level ？ Not at all , It's just that the probability is 1.2 Near the level of .

in addition , The operation process inside machine learning is also full of randomness , Especially the major algorithms based on probability graph model , This section begins , Avoid introducing obscure concepts for the time being .

Therefore, the randomness of machine learning in programming is different from the certainty of traditional programming .

2 Machine learning and AI The relationship between

2.1 AI three University School

Throughout the history of AI development , There are three mainstream academic research schools .

• Semiotic school
  The school of semiotics believes that ： Any system that can operate some patterns or symbols and convert them into other patterns or symbols may produce intelligent behavior . The core of symbolism is mathematical logic , It focuses on the advanced behavior of human intelligence , Like reasoning 、 planning 、 Knowledge representation, etc , It is committed to simulating human cognitive process and abstract logical thinking with computer symbolic operation .
• The connectionist school
  The connectionist school believes that ： Trillions of neurons in the human brain are intricately interconnected , It is the source of intelligence . The core of connectionism is bionics and neuroscience , It focuses on the connection mechanism and learning algorithm between neural networks , Dedicated to representing a large number of neurons through computers , To simulate the intelligence of the brain .
• Behaviorism school
  Behaviorism school believes that ： Intelligent behavior is used by individuals to adapt to environmental changes 、 A combination of various physical reactions . The core of behaviorism is control theory , The focus is on a kind of “ perception - action ” Behavior intelligent simulation method , Committed to anticipating and controlling behavior .

2.2 Machine learning is in AI in

Artificial intelligence is a very large concept , It involves a lot of content ： Computer vision 、 natural language processing 、 machine learning 、 Deep learning 、 Robot control and so on , These contents will also cross each other , Involve each other , in general , Machine learning is a branch of artificial intelligence

Deep learning takes neural network as the core , Neural network is one of the basic models of machine learning , Simple understanding , Deep learning is machine learning that stacks many layers of Networks .

3 What can machine learning do ？

Machine learning can be applied in a wide range of fields , Such as medical treatment and health 、 Business services 、 Consumer retail 、 Finance 、 robot 、 Agriculture 、 education 、 Lawyer legal affairs 、 Social media 、 Automobile traffic 、 Smart home 、 Communication service 、 Urban public 、 Tourism and other fields .

• Search engine
• Recommendation system ( What does Tiktok push you , Why can your article be on the hot list )
• Aid decision making
• Medical robot
• Smart wear
• …

4 Learning route

It takes so long to introduce the concept and application of machine learning , Next is the most important , How to learn machine learning ？

We are positioned to engage in innovative work , Instead of repeating the previous models . Do that , You must have a thorough understanding of the principle of the model , Only in this way can we know what defects the model has , How to improve , How to create your own model .

This is also Machine learning strong foundation program The purpose of , We explore the mathematical principles behind the model , To see the advantages and disadvantages of the model in essence , What are the constraints , What is the idea of model improvement , Try not to stay on the surface .

• The first stage ： Basic tools
  Master a programming language , Recommended to python introduction . Machine learning as an applied discipline , Finally, it must be implemented into programming . The derivation of machine learning algorithm requires a certain mathematical foundation , There are about three directions ：

  • probability theory
  • linear algebra
  • optimization

  Some models also involve graph theory 、 Some theories of functional analysis and other disciplines . Those who don't know can first understand the basic concepts of these three mathematics , Or learn from what you use in this column . The level of mathematics learning is not high , Just use mathematical theorems to prove machine learning theory , There is no need to delve into the origin of these mathematical theorems .

• The second stage ： Basic model of machine learning
  Learning machine learning models at the theoretical level , Standing on the shoulders of giants , Understand the thoughts of predecessors and the basic problems faced by machine learning

  • How to divide data sets
  • What are the relationships and differences between models ？
  • How to improve model generalization ability ？
  • How to select a model according to the scene
  • …

  Three textbooks are recommended for machine learning theory ：《Deep Learning》、《 Statistical machine learning expericnce 》、《 machine learning zhou 》

• The third stage ： Combined with specific application directions

This column is based on the first stage , Positioning in the second stage , Try to explain clearly the unimaginable details in the textbook .

---

More wonderful Columns ：

• 《ROS From entry to mastery 》
• 《 Robot principle and technology 》
• 《 Machine learning strong foundation program 》
• 《 Computer vision course 》
• …

The source code for · Technical communication · Tug of learning · Consultation and sharing Please contact the