[Highly recommended] How to do scientific research

Main content: Mainly summarize the scientific research experience of a student I like very much.
I listened to a lecture by Professor Zhang Zhenjie yesterday. For me, a scientific research novice, it is necessary to make a summary. Although many principles are not deep, they are still You can think immediately, and prepare for making fewer mistakes and lower-level mistakes in future research work!

Combine the professor's explanation and my understanding. Summarized as follows

What is good scientific research:

---

From the Dao to the Jane, there is a summative scientific research work: to discover and reveal the laws of things in a minimalist way. It must be a good job. Li Xiaowen, an academician of cloth shoes, once said, “Science itself should pursue the principle of simplicity, and the simpler the better.

Leading the direction of highly enlightening scientific research: a job is enough to "instruct the world and create a world", just like the deep learning GAN proposed in 2014, and since then, GAN has become more and more popular, the reason is GAN His thoughts are very enlightening and innovative!

Filling in the gaps and practical scientific research: discovering important laws that were not discovered by the predecessors. What needs to be emphasized is that not all scientific research that fills in the blanks is good work, and some work is meaningless, that is, work that has no practicality.

Topic selection

---

New bottle or new wine? If the new bottle represents the research object, new problems in the research field, new hardware, and new applications. New wine refers to new research methods, such as: new algorithms, new analyses, and new implementations. When choosing your own subject, how should you weigh it?

The first is the bottle, that is, how new the research field needs to be? When entering a new field too early, you should understand: 1. Will this topic become popular? 2. How long will this topic be popular? And when it is too late When entering a field, you should know: 1. Is there room for improvement in this field? 2. Can the remaining problems be solved?

Several ideas for finding new bottles or new wines:

---

Application expansion: When looking for an idea, you can look for it from an application-driven perspective. For example: when the social network just started in 2005, J Leskovec and others seized the opportunity and made the first social network Research on construction on the network. Published Graphs over time: densification laws, shrinking diameters and possible explanations. This is to find the idea from the perspective of the application. When you need to find inspiration for applications, you need to understand the following questions: 1. Is it necessary for practical applications, 2. There is no ready-made solution, 3. How important is the application?

Applicability expansion: This refers to when we need to find an idea, we need to think about whether the previous method has its limitations, and where are its limitations? After finding its applicable limitations, try to expand its applicability, but at the same time, it is necessary to figure out whether such an extension is necessary.

Dimensional expansion: In my opinion, dimensionality refers to the dimension of the field. Dimensional expansion is the expansion in the field, that is, when external factors change, such as the current cloud computing and the development of new hardware GPUs, can we? Take the problems that the previous algorithm could not handle. For example, in Mars: a MapReduce framework on graphics processors, when the GUP was just on fire, MapReduce was put on the GPU for calculation.

Cross expansion: This is whether we can combine several concepts to do it. Of course, at this time, it is necessary to solve the problem of whether the application scope is too narrow.

In short: we have to read more articles in different fields to understand the progress in different fields and the progress in the industry. Of course, now that the information is so developed, we can easily obtain such knowledge, such as through the official account, Zhihu, etc.!

1

Research methods

As shown above: not every job necessarily requires innovative technology to be used in new problems. Most successful jobs are a combination of (new bottle, old wine) or (old bottle, new wine). In our research, it means that it is not necessary for a new bottle to be a new wine. It is more about finding a new method on a problem, or using a method to solve a new problem!
The picture below is a sentence of Einstein, which is worth thinking about by researchers:

How to think about the above picture, there is an angle that we must learn to tolerate failure. Although this principle is simple, it is true for us people, especially those engaged in scientific research.

Then we will definitely ask what to do if the method fails? To sum up three points:

1. Understand the reasons for failure, algorithms, implementations, assumptions, where are the problems?
2. After this method fails, can we change the problem, how to find the problem, and return to the above method such as cross-expansion, such as adding additional assumptions.
3. Finally, let it go if it's time to let it go. Although this is really difficult, as the saying goes, your own idea and your own work are as precious as your own children, and it is not easy to give up easily. But letting go in time is also a kind of wisdom!

In addition, do not do too much Straighforward in scientific research, because there are many uncertainties in scientific research. For example, for computers, if you know what you can do at a certain time, then it is not called scientific research, it should be called development!

Finally, when discussing with our tutors or classmates, we should do more self-criticism. Many people will talk about the correctness of their own methods when discussing. Of course, perhaps scientists are a little bit paranoid. Resolutely consider the feasibility of your own method. But on the other hand, when you have disagreements with others, you really have to do more of your own reflection. At the same time, I expand my thinking, from research to innovation.

2

Publication of papers

When publishing a paper, first consider where to invest? Must top conferences and journals? The answer is: yes. Second- and third-rate conferences are really like drugs. They can make you full of confidence, but also make your scientific research level low. When you have been posting second- and third-rate meetings for a long time, you can only be second- and third-rate meetings.

Of course, how to know whether your job is a category A job should be discussed with your boss. I think the most important thing is that mentality. It is necessary to have a mentality of pursuing the ultimate, to be the best mentality, instead of stopping after a little achievement and starting to publish papers for satisfaction.

Then after submitting the manuscript, it is very likely to be rejected... This time combined with the reviewer’s comments, of course, those comments may not be true, because the reviewer will only give him the safest reason for rejecting you. Not necessarily the real reason.

For example, your assumption was wrong at the beginning, and the whole paper is really very watery in the eyes of the reviewer, but he may not say that your work is too bad, but may say that your experiment is not well done. Reject, because he also worried that he was wrong, and the safest way to reject is to use some common faults or defects. So when looking at the reviewer's comments, look at it objectively. If your paper is rejected, you have to figure out whether it is an experimental problem, a hypothetical problem, a data problem, or even a writing problem!

Does writing really matter? Yes, writing is really important. When we write an English paper, we must write clearly and clearly. As shown in the picture above, the first impression of the above example is that the expression is not clear. The Chinese will not be directly translated into will. It is inappropriate to use the future tense here.

The above example is in terms of variable naming. It is better to change the ps to an English letter or a Latin letter! Of course, this is not something you can practice in one or two days, you need to practice more.

3

Follow-up development

If your paper is accepted, this is a good time to promote your work when attending conferences or other exchange meetings. When giving a speech, you should pay attention to less routines. Don't sell for the sake of sales, and put your work skills in too much detail, but more motivation. Talk more about how your thoughts came about and how you think. It is also very enlightening to the listeners, and they are also willing to listen to your speech!

When attending various conferences, communicate more with colleagues and pay attention to posters, because many people look at the posters first when going to the meeting, so we can only promote our work better if we make the posters well.

4

to sum up

How to find a good problem, a good way, how to know how your job is: This is to discuss it with your boss.
How to think of a good way to find innovative points: this one should read more different papers, and use more critical thinking to think about the problem!

How to write a good article: This is to write more, only familiar with the hand.
How to make the method practical: Join the open source community and open source your own work.
What kind of people can really do scientific research: In the words of the boss, those who are self-driven, those who are really hardworking and thinking!

Recommended reading:

Selected dry goods|Summary of dry goods catalog in the past six months
[optimization] basics of linear programming
[intuitive detailed explanation] what is PCA and SVD

      欢迎关注公众号学习交流~         

Welcome to join the exchange group to exchange learning~