[note] the art of research (understand the importance of the problem)

3. Understand the importance of the problem

How to find the problem to study
Think clearly about the importance of your research

Finding problems to solve is a unique thing in the research field , It is very important to think clearly about the meaning of a thing before doing it .

3.1 FROM AN INTEREST TO TOPIC

In their own interests , Find a topic , Then adjust its size moderately , So that we can do research on it . For this topic , We need to ask some questions to know that we do research .

Some definitions ：

• subject： Academic field , Like computer vision 、 Artificial intelligence, etc ;
• topic： topic of conversation , That is, special points of interest in the field , For example, how to efficiently design convolutional neural network to make the classification accuracy of pictures higher .

Why choose a topic ？
Because it's hard to do a research on the whole research field , So you must find a topic in your field , Only in this way can we dig deeper .
Topic is a way for you to ask some questions in it , If you answer this question well , Researchers in the whole field will be interested in it , And a good answer can promote the development of the whole field to a certain extent .

The topic itself can be larger or smaller .

• Big means , A common topic in the whole research neighborhood
• A smaller one is also ok , But only certain researchers are interested
• There must be more people involved in the research on topics that everyone is interested in , So the competition will be more intense
• The probability of doing a better job than others is higher when it comes to niche topics , But it will make your work less influential , The audience is not so wide
  The most important thing is to ask some questions about this topic , Then the answer to your question , Others may feel important , Even this answer can change the whole research field to a certain extent .

Distinguish what is question and problem？

• question Is doubt , Ask a question , Then you need to answer ;
• problem It can be considered a difficulty , There is a problem , If you don't solve it , It may bring some harm

How to change from interest to a topic ？
Find your interest after reading some articles , Then find a topic . For the technical field , There are four main things to consider ：

1. How to do it （ Not before , Now you can ）
2. How to expand （ data 、 scale ）
3. How to make it cheaper
4. How to do it safely

3.2 FROM A BROAD TOPIC TO A FOCUSED ONE

How to make the topic smaller , And enable it to start doing research ？

How to make the topic contribute to a better research work ？

Can you turn this topic into an argument ？ This argument seems to be worth discussing .

3.3 FROM A FOCUSED ONE TO A QUETION

When you find the right topic , It's time to ask some questions . While reading the paper , It's best to bring some specific questions （ Can Gu really become some of the final research problems ） Come and read , Concentration will be better .

3.4 THE MOST SIGNIFICANT QUESTION: SO WHAT?

When interested in a problem , You should ask yourself a more difficult question ,SO WHAT ?

• If this problem is solved , So what ？ Will others care ？ What benefits can it bring to others ？ Whether it can promote the development of the whole field ？ Or promote the development of the field in a small place ？ Can you inspire others to do more work ？
• You can also ask the opposite , If you don't do this problem , What kind of loss will there be ？ Will others feel that they have missed a hundred million ？ Is it possible that the whole field will stop developing for many years ？
• If your answer is no , That is to say, it doesn't matter to do it because of interest , This allows us to go further , But keep asking SO WHAT This problem , Think about what I have now SO WHAT, others care No care.

Find your topic + Ask a question + Explain the importance of the problem

TOPIC: I am studying/working on ________________
QUESTION: because I want to find out who/what/when/where/whether/why/how _____
SIGNIFICANCE: in order to _________________
reflection ： Why does this matter ?

QUICK TOPIC: HOW TO FIND A TOPIC ?

1. Browse through a textbook of a course that is one level beyond yours
2. Attend a lecture for an advanced class in your field
3. mentor
4. Look online and talk about
5. …

The core idea of this section ：

• How to find a topic from the point of interest , Then narrow the topic down to a small enough scope , So that we can control it , It also makes it big enough to support a better research ;
• There are many questions to ask on this topic , Because our research is used to solve problems , So there must be a problem .
• When there's a problem , You have to ask , The most important so what The problem of , That is to say, how to solve this problem . Before really starting to solve the problem , I really need to think about it , Suppose half a year later ,1 Years later, I solved this problem , So do others care about this matter .
• Before doing any project , We should think about the significance of making this thing , Don't think about whether you can do it . Just assume that there are enough resources , Good luck enough , It can be done with the best results and when it can be done , Think about the impact on others （ Where is the ceiling of the meaning of this matter ）.

4. FROM Questions to a Problem

4.1 Understanding research problems Understand research questions

There are two kinds of problems to be studied ：

1. Practical problems （practical problem）： For example, an algorithm runs slowly , Solving him will make it run faster ; The accuracy of a task on a certain data set is not very high , We can propose a new algorithm to solve it ; There is no good data set in this field that can make you better evaluate this problem , Then I propose a data set ; This is to solve some practical problems

2. Conceptual problems （conceptual problem）： These problems are not to solve a practical pain point , But if we answer him , Will have a better understanding of something or the world . In the field of Science , Most of the problems are some problems in this concept , Help us better understand something .

   But in technology , More practical problems . Of course, there are many theoretical problems , For example, understanding batch normalization Why can work , And why deep learning can work , Even why machine learning works . There is a lot of theoretical work , But we are only so far , There are not many papers dealing with these , So you may feel that these things are not much . But in fact, in the whole field of artificial intelligence Especially machine learning , Maybe before deep learning , There may be at least 1/3 My paper is about some conceptual problems like this .

4.2 Understanding the common structure of problems

Why should there be a problem ？

If you have no problem , Maybe your whole paper is written with a lot of data in it , A lot of evidence is in it , But others don't seem to be able to find your point of view （point） Where is the ;
Many novice writers just don't focus on a problem , Make readers can't keep up with your rhythm , I don't know what question I'm asking , So there is still a problem .

What is the structure of a problem ？
Both practical and conceptual problems have a following structure ：
situation condition That means your condition
A result you don't want （ If we don't solve it , You may need to pay a price you don't want to pay ）
Because your situation is different from your price , The two kinds of problems are different .
What needs attention is the so-called cost , In fact, readers are concerned about , It's not the price we have to pay , It's the price they have to pay .
How to find out such a cost from the perspective of readers ？ Start from your own situation and keep asking SO WHAT .

4.3 Finding a good research problem

Will readers think it should be solved ?
Readers: Not I don't agree, but I don't care.

How to find a good research problem ？

• Ask for help ： Discuss with others and keep asking SO WHAT The problem of ;
• Look for problems in reading ;
• Choose from your own summary , When writing, I usually think more comprehensively , So it's written objectively , Then when we write it down, it also helps us think more comprehensively ;
• Finding a good problem is a thing that a researcher may spend many years , We need to cooperate with researchers who may be better at finding good problems than us , Learn from him how to find good problems in this field .

How to get rid of the anxiety of not knowing whether the problem is important when entering a new field ？

Don't wait until the research is really finished 、 The thesis is finished , Then show others whether this thing is worth doing . At the beginning , When you have an idea , Talk to others , With my classmates 、 colleagues 、 teacher 、 Even for users who may be targeted , See if they care about this matter .

Don't wait until your research is really finished , It can be done 1/3 perhaps 1/2 When , Prepare one 10-20 Minute report , Tell you about it in some research groups , Take a look at your feedback .（ Can be in workshop Submitted above , Look at everyone's thoughts ）

At work , When you want to do any research , Before doing it, you must make it clear which product team or customer may need this thing , Will they put this research work into the product . Don't rush to do it before doing it , First think about what kind of problem you want to do , What methods may be used , Then you talk with these possible partners in the future , Tell him ： I'm going to do this problem , Are you interested , Are you interested in putting this thing into your product .

The core is to do research , Keep talking to potential readers （ user ） communicate , This interaction can help us change one direction of the whole research . It is also to ensure that after the research is completed , Finally, someone cares and someone buys .

5. From Problems to Sources

Look at the previous work cited , Quote its later work .

In the field of Technology , To solve a problem , First, find oneortwo papers that are most relevant to your problem , Then go to see who he quotes , Just walk forward , Look at what his previous job looked like .

Don't read a paper summarizing the previous work , Go and take a look at the things he really quotes , Complete a summary of yourself , Then let's see who quotes this article ; Move this paper forward and backward , Basically, we can find the resources we need .

6. Engaging Sources

How to evaluate a paper ：

• The papers of twoorthree years are quoted , The source of the new paper . Look at the quote , Probably know how many people read this article , How many people care about it , This method is mainly aimed at articles that have been published for several years ;
• For the new article is to say , At least in the computer world , Both magazines and conferences have rankings , Although the ranking is not so credible . Suppose the main articles cited in an article , That is to say, our method 、 Those articles that make comparisons or are based on , At a meeting , Our articles are usually published in the past ;
• Want to publish it at a first-class meeting , The cited papers should also be articles of first-class conferences or journals . Research is also in circles , Suppose you want to post an article about a first-class meeting , Then we read and quote , And based on the best work is also a first-class meeting, or even better .
  First class meetings ：NeurIPS,ICML,ICLR … CVPR（ Partial application ）

Reference: B standing Learn from Li Mu AI

---

$$Author:Chier$$