You want to publish a paper or do scientific research, but your supervisor doesn’t care about you, and no one takes you, what should you do? How to get started?
I will divide into 3 sections and tell you how to solve this problem.
1. What role should a mentor play in scientific research?
1. What is the whole process of scientific research?
The first piece, everyone must first have an overall understanding of scientific research . What is the whole process of scientific research like? You first need to choose a research question, that is, a topic;
Then the second step is to investigate the existing solutions to this problem , and the third step is to start based on their solutions. You have to think about your own idea . After you have an idea, you need to do experiments to verify it and fully verify it. , to prove that your idea is effective and outstanding enough;
After this step, the next step is to write . Writing is to express your idea, experimental results, and methods in a very detailed form, and submit it to peers for review to see if they accept the article. If accepted, congratulations, you have your first paper.
This is the whole process of scientific research.
But this process is actually the most ideal situation-go through the process once, and the paper will be sent out. And 99% of the cases are more likely - the first idea you think is unreliable, and your experiment cannot verify the effectiveness of your idea .
What do you need to do at this time? You have to continue to think about ideas, rethink new ideas, and then do new research again, just like this over and over again, a whole cycle. It may be after many iterations that you can verify an effective idea. After you have an effective idea, you can actually start writing the paper, and then there will be subsequent submissions, reviewing, and finally the final mid-draft. This is the overall process of a process.
Then the question arises, what problems will you encounter after this set of scientific research process is completed? What problems will all the students who do scientific research encounter? Or what capabilities do you need? How do each of these steps need to be done to make it happen? This involves some corresponding things in each of the following modules.
The topic you choose is either given to you by your tutor, or you combine your own interests. The same is true during research. You have to read papers intensively, read papers extensively, and sort out the research context; what about the stage of thinking about ideas? In the process of thinking about ideas, many students will get stuck for a long time, but in fact it has some methodology and thinking methodology, which will help you to clarify the direction of thinking. At the same time, during this process, some of your logical thinking and criticism Sexual thinking will help you to verify your idea. When you think, you will continue to overthrow it, and then think about new ones. Just like this, you will polish your idea repeatedly. In the end, your idea will be a very new one. idea, others will have comments like that's interesting, that's novel.
Of course, you can’t just think about an idea in a wild way, you have to do experiments to verify it. So how to design experiments based on idea? How to implement it with code? How to debug it? If your experiment runs through, the code runs through, and the experiment is finished, how do you analyze the results of your experiment? There are many links in between.
You don’t start to face the problem of writing until after all the previous links are completed. Although writing is the last thing, writing is very, very important. Your writing directly determines whether your article can be accepted in the end. I have personal experience, the same thing, I changed a different way of writing, the first way of writing was directly rejected, others said "your work is not innovative", the second way of writing, others (reviewers) accepted up. So how to write a dissertation is actually very deep inside, and there are many skills and experiences to talk about. Specifically, I can summarize it like this:
tell a story. How to tell a story? Draw out your idea reasonably, let others accept it, let others think it is reasonable, it is good, and recognize it, including the production of diagrams and tables in your paper;
Then there is the reply to the review comments . The reviewers will ask some questions after reading your paper. How do you answer these questions? How to get him to recognize your work? There is a lot of experience in all these aspects.
Next, let's think about it from a more macro perspective.
First the first part is the input. We divide the entire scientific research process into three parts, starting from selecting a problem, and then reading a variety of literature, whether it is intensive reading or extensive reading, this is a continuous input, and these are in your brain. Absorbing all the information and knowledge is only the first step; in the process of thinking about ideas later, verifying ideas with experiments, and continuously circulating, after absorbing them, you have to think, understand, and then transform them into your own ideas. You need to do experiments to verify it. When you verify and analyze the experimental results, when you design experiments, write codes, and debug, you are constantly thinking about whether your idea is feasible; The last step is output, in the form of a systematic academic paper, presenting all your ideas, methods, and experimental results, forming a standardized paper that can be recognized and accepted by peers. So it has three levels—input→absorption, thinking, understanding→output. This is the entire scientific research process.
2. What role can the mentor play in this process?
Then everyone can think about a question, what role do you think the mentor should play in this process? In other words, in what ways can your current mentor help you? For example, in the five steps of 12345 in this picture, in what areas can your mentor give you help and guidance that you find useful?
Why can tutors participate in all aspects? And can help you?
If a real mentor participates in the whole process, how much guidance can he give you?
First of all, when choosing a topic , the instructor's experience judgment is very, very, very important. If you go in the wrong direction at the beginning, you will only step on the pit later, and you don't know how long this pit will last. A qualified tutor or a tutor who is responsible for the students will tell you at this time, on the premise that you have no foundation and no knowledge, will help you judge some reliable directions, and then tell you what to do where to go. This is the first step to choose a topic, and you need to rely on the experience and judgment of your instructor;
Second, it’s the same when researching the project . Since the supervisor specifies this topic for you, he will not specify it blindly. He must have a knowledge and grasp of this field, so when you go to research, he may Give you some very classic, very cutting-edge, or very important papers, then at this time you just need to read the papers honestly according to his guidance, whether it is intensive reading or extensive reading, just follow Read his guide, and he will help you sort out the context of the research plan. In this process, you will learn the basics, and he will tell you regularly to see where you are now and how you are progressing;
The process of thinking about an idea is the same . He will guide you and let you think about it. After you have some ideas, he will tell you whether the idea is reliable or not, and whether you should think about it from there. If you have some problems in the process of thinking, he will point them out to you directly, and you will save a lot of detours. If your idea has some problems at the theoretical level from the beginning, if no one tells you, you will do experiments, prepare data sets, type codes, run experiments, and find out that everything is wrong after running an analysis. It is a bad result. At this time, you will wonder whether I did not do the experiment well or the idea itself has problems. You may continue to do other experiments. At this time, if there is a qualified mentor who tells you directly from the beginning that this idea is unreliable and cannot be done, and you think about new ones, then you will save a lot of time and avoid many detours .
This is why I said that when thinking about an idea, it is so important for the mentor to pick out some of your problems and point out some directions for you, and then go to the next step to experiment and verify the idea, he will analyze, now your idea is If it is effective, it is more reasonable and reliable. Then you have to verify it and design the experiment. He will guide you to ensure that your experimental design is reasonable and comprehensive, which is also very important. At the beginning, if you design the experiment yourself, you will miss a lot of details, and you will not be able to figure out how to do many experiments, but the instructor is very experienced.
The only two cases I have heard of are the only two cases I have heard. One is a doctoral student in computer science at Purdue University in the United States. I posted a circle of friends; another example is in China, a young teacher is the same, he will teach students to write code, just in his office, directly see how to write with his students. I have been doing scientific research for so long now, and I have only heard of these two tutors who can lead students to write code. So in fact, there are, not none, and will take you to analyze the experimental results, let alone, since you have the results, you must analyze them, and then determine whether the idea is reliable or not. In the middle, the instructor is also It can take you and guide you to think, analyze, and write until the last step.
In terms of writing, the tutor will usually help you write, because he thinks that your thesis is about to come out, and the experiments are all done and effective. Let’s write the thesis happily. My student is about to win a thesis. , I will take you to do it. Therefore, if a general mentor is willing to spend time, he will spend it in the last step, because the time ahead will be very long, and his input-output ratio is actually not high. But if you don't help you before, these steps of 1234 will take a very, very long time. For students with zero foundation, the tutor actually wants to help you write and draw at the beginning, and also help you answer review comments, help you draw pictures, make tables, and some will write papers for you. Some teachers will write your paper from beginning to end, especially for beginners. The mentor guides a novice who just has a little research experience and can only do experiments and think of ideas. If you ask him to show his research results in a systematic way, he can’t write them out, which will affect the quality of his final paper submission for ideas.
At this point, we should have a good idea of what role a mentor can play in scientific research as a whole.
From my experience, in fact, the relationship between our current students and their supervisors, or how everyone is doing scientific research, is actually the most basic "teacher-apprentice system", also called "apprenticeship system " . What does "Apprenticeship" mean, I posted the introduction on Baidu Encyclopedia directly in the above picture, in fact, it is in the form of "taking you to do it". Take scientific research as an example. Your "apprentice" acquires knowledge and skills under the guidance of the master. It is a highly contextualized learning method. Capture the knowledge and skills of the master, practice under the guidance of the master, and gradually learn the skills of the master."
The same goes for the skills of scientific research. I have just split them up. There are so many pieces to learn. During this process, he needs to continue to lead you to do them. Only in the real scene can you learn a little bit. Ascension, this is the most basic. This may also be more common in foreign countries, that is, teachers will guide students in this way and lead them to do it. This is to train students and give scientific research guidance.
There is another kind that we all know what it is, the "group army". The "group army" is this kind of large scientific research team. It has a large amount of computing resources and a large number of interdisciplinary talents. For example, deepmind, meta reality lab, Facebook before, and BAT now have some research institutes. Like the Bodhidharma Institute, they are now fighting as a group army. No one fights alone. They all work together to publish some high-quality and influential articles. This is the way they do scientific research.
Then what about us, the situation we are encountering now is single-soldier combat, and I, as a single soldier, have to do everything, so who am I at this time, where am I, and what am I going to do? what should I do? A bunch of questions came out... This is really a helpless thing, you are facing a huge subject, and now you are a new student who knows nothing, everyone is just getting started No, it’s reasonable, but you have to face such a huge problem. Is it realistic for you to challenge an army of others by yourself? It may be made by several people working together, but you are alone, how can you do better than others? How can you publish articles of the same level as others?
So here is the problem at this time, why is this happening? Where is the problem?
Why does this happen to so many students now, but they just don't know what to do? The tutor doesn't care about you either.
2. The whole process of issuing CVPR from scratch
The second part enters the key content of this article- the whole process of sending CVPR from scratch .
■ Where does Idea come from?
■ How to verify the experiment?
■ How can I not lose my mind after being rejected 3 times?
If you look back at the state of mind when I published the paper, it was the kind of ambition with a heart higher than the sky. I will publish the top meeting, I will publish CVPR, and none of my labs will publish it, but I will publish it... I Let me tell you how difficult this process was, and how it was finally issued.
Let's take a look at the overall time period first. If you want to post a CVPR, how long do you have to prepare?
In this entire cycle, the first three modules actually took me 4 to 5 months, starting from scratch. Because when I first entered the laboratory, I actually only knew some basics of programming, because I only knew some computer databases and so on as an undergraduate, and I was not very good at mathematics, basically zero foundation. Then the previous question was given to you by your instructor, so you can follow this and use this to select a topic. After you finish the research, you have to do research. During the research process, you have to learn various things and think about some things. Ideas that may or may not be reliable will take 4 to 5 months.
Then in the middle and late period, I was inspired once. When the idea came out, I did some simple experiments and found that it worked, it completely met my expectations, and there was really that kind of relationship, and the value really just matched. My expectations are consistent with my expectations. Finally, I had a reliable idea, and I was very excited when I got the experimental results. After you have this, you can start to arrange various experiments, and the rest is for you to improve your effect on this experimental method, and then adjust the parameters...
In fact, these two months have passed quickly, because you will do a lot of experiments, and the results of the experiments will tell you every day, whether your ideas are OK, what are the specific details, and how should you design them? It only takes two months. I verified that the idea actually took only two months, but I think it took 4 to 5 months for the idea. It is a process of accumulation and accumulation. You will spend a lot of time in front of you and you will not see feedback, but after that After that node, you suddenly find that there is some feedback and some good results.
Then came the writing, and it took two or three months to write, and various things happened in the middle.
1. Where do ideas come from?
First, how did I find my idea?
This first has to go back to the problem I want to solve. The problem I want to solve is that I need to train a segmented model with one month's data to the next domain to transfer this knowledge point.
I have already got the problem I want to study, which is to migrate the segmentation model in a clear scene to a foggy and harsh scene, so how to solve this problem? In other words, what are the current research programs for these issues? After doing some research on my own, I found that there are actually very few existing solutions. I tried to reproduce various solutions and found that there are not so many effective solutions, and the effect is average. At this time, I have a question: how can I solve this problem? Is this problem solved? Can it be solved?
I just read a lot of papers, others are doing this field, but what about other fields? Are there any similar fields? How did those people do it? Or to track down some of the latest papers, is anyone doing something similar? Then I read it every day. One day I saw a person on Zhihu who shared the thesis he had just won. It hadn’t been published yet, but he had already won it. He found that his subject was very related to mine, so he did it. For one thing, he threw a picture into a segmentation network, and two feature maps were obtained from different branches of the segmentation network. He calculated the similarity of these two feature maps, and he found that the picture with this similarity was actually the same as the last The segmentation results are consistent with the difference between the real labels.
It is equivalent to him telling you something in this way? Let me tell you, my network can actually tell you some things that he may not have learned well. After you have this information, you can improve this information in a targeted manner. I tried to find a way to do it well, and then the network performance was brought up. That was an inspiration for me at the time. Now I will talk about this, because everyone’s knowledge background is not consistent, and not everyone can understand it. But you know that I read that paper when he hadn’t published it yet, but the author had already told him that he had won, so I went to read it, but he still published it, so I ran away and ran around. , I found that it is very effective for my task, this is the first step.
2. How to verify the experiment?
As for the details later, I further put forward some ideas of my own at that time, that is, if I go directly from this source domain to this target, the domain gap between the two domains is very large, and my I always thought, can there be an intermediate domain? The middle domain is equivalent to a bridge that connects these two domains. There are some clear scenes between the two domains. The difference between them, I regard it as the difference in style. Both of them were taken in the same city, and the difference between them is only the difference of fog. Turn this large domain difference into two small domain differences, and then we use some inspiration from the paper in the previous place, I used this value of MVV. This value represents the difference between the models that can be measured, the gap between different domains, and how big the gap is. Then I was surprised to find that from this perspective, this The two domains can be distinguished, the two gaps can be distinguished, and they have an additive relationship.
At this point, my idea is basically verified. I have been able to prove that what I think is right, and there are experiments to prove this result, then I will design my specific method, and then I will start writing the paper after the method is designed.
Then what was the division of labor in writing for this kind of DDL that I met for the first time, because my mentor didn’t care about me at the time, I found a senior by myself, and then I was in the same direction as me. Well, it’s not quite the same, but he has some experience anyway, so I told him, let’s do it together, and now I have some results, but I don’t know how to write a thesis, and I haven’t experienced the following things, then He just said to cooperate, take me together, and I will start to do this. It is the first time for us to catch up with the DDL of the top meeting. It is the first time for me. You can see that the figures and tables of my thesis are wet. Brother comics, the text part of the thesis, except the method is written by myself, because my method is the clearest, the rest of the parts are written by him, and then all the experiments are done by myself, and then include registration and submission later, various All the precautions are done by myself, and then I will give feedback to my brother every time, and then see if his opinion is right, so the first time our submission is over, we will hurry up on that paper Hurry up, the one or two weeks before the final submission of the manuscript is simply crazy. Maybe one day until the last week, you may sleep for 3 or 4 hours a day, and then spend the rest of the time in the laboratory, sitting in front of the computer and writing papers , do experiments.
3. How can I not lose my mind after being rejected 3 times?
Novice fighting monsters, full of expectations, ushered in our first submission. At that time, it was January 2021, and IJCAI was also a top meeting.
Our version at that time probably looked like this, and the intro looked like this.
Then the method diagram is drawn like this.
Then include the details of some specific methods, some loss functions, some algorithm flow charts, it looks like this.
After we get here, let's take a look at its score after we submit it. After more than a month, there will be a score, so the score is 66654.
I was fortunate enough to meet five reviewers. This score of 6 is actually a pretty good score, which is equivalent to a weak reception, which is a score above the average line, and then a score of 4 is a score below the average line. Looking at the overall score, if there is no 4 points, it is actually not bad. After having the 4 points, it is actually not very optimistic.
At that time, I read a lot of comments from reviewers, and they were pretty good, but I had no experience, and I didn’t know if what they said was really good or what. Finally, after we got the review comments, we specially responded to them. Look at the picture above, five reviewers, our motivation ... is interesting and insightful. Our proposed method is interesting, novel and effective. Anyway, it’s just a meal Exaggeration... In fact, every reviewer mentioned some advantages, and then said that our avlation study is interesting and experiments are sufficient , the experiment is very sufficient, your experiment is very interesting, and I also made a detailed one Analysis and comparison, compared 15 methods... and so on and so on.
But what about the result? It was still rejected. We replied after the rebuttle was over. It has been more than a month, and the rejection is written like this: Unfortunately, I regret to inform you that your paper has been rejected. This is our first submission, and it was rejected.
The first time is over, so we definitely can't let it go, revise the draft with tears, and continue to improve! Continue to fight and continue to vote, and it will be the second time.
The second time was in April 21. In April last year, we made a lot of improvements this time. The method and the intro diagram were also changed, including the motivation diagram. In fact, all of them were redrawn, and the method diagram was also redrawn. I drew it again, and then these patterns and processes, we feel that they have been improved a lot, including doing more experiments than before, and adding some forms, and then the results - borderline reject, weak reject, borderline There is only one good accept, and the remaining two are very bad. At that time, we were all shocked, and felt that I had made so many improvements, and the score was still not as good as the first score. It was really embarrassing! I don't want to mention the reviewers' comments, because I think they are very unprofessional...
After reaching this step, we were rejected for the second time, so we will continue to vote for the third time, continue to change votes, and become more courageous with frustration. The third time has been in September of 21, that is, last year, probably in September of AAAI, then I will continue to submit, and we have actually made some improvements when submitting, including these intros, diagrams, including some method parts , The diagrams of the method have also been updated, and a new version has been drawn, including experiments that have been done more fully than before.
The result is reject again.
AAAI is divided into two stages of review. In the first stage, you will be divided into two reviewers. If the average scores of these two reviewers are relatively high, for example, both of them are weak accept or borderline accept. He can give you a chance to enter the second round if it is above the receiving average, otherwise you will be rejected in the first round. We were lucky, we were rejected in the first round, a direct reject, a very low score, and then a borderline reject, both reviewers were dissatisfied, they thought, "what are you writing?" ? After more than a month, the results came out, and we were rejected... Is this luck or misfortune?
At that time, my mentality was finally on the verge of collapse, and what was the only good thing? Another article of mine entered the second round, which gave me a little comfort, and it would not be so collapsed. Of course, that article actually It has already been selected, and this article has been selected for the third time by me, and it has been changed so much, it has been improved a lot, and each version has improved a lot on the basis of the previous version.
I voted for the fourth time, and for the fourth time I voted for CVPR. Is it time to win? It has been almost a year. This year, from IJCAI's 2021, to MM's 2021, to AAAI's 2022, and then to CVPR's 2022, there have been four times!
It’s November of last year, and our method has changed from the previous AAAI to this, and we have done a lot of experiments, because he attacked you before, saying that your experiment is not compared with this method, Why is there no comparison with that method? We have done all the experiments now, including all the available methods on the market, including these experiments are also done very fully.
Then let's look at the review comments. The first three lines belong to one person, the middle one belongs to the second person, and the bottom one belongs to the third person. This is the later CVPR review comments.
Finally met some reviewers who know the goods. Of course, we are also constantly improving our papers. After this step, you will find that it seems to have played a little role. The comments of the reviewers are very positive, and I have intercepted them all. , Of course, many questions will be asked in it, it is impossible to just some positive opinions.
After he asked the question, we had to reply. After writing a version to reply to the review comments, our score was raised by one point. It was three weak accepts before, but after writing the review comments, it became Two weak accepts and one strong accept.
The picture above is a comment written by a reviewer after reading our rebuttal reply to the review comments:
He said that he liked the author's re-doing of these new experiments to analyze and discuss how the intermediate domain was generated, and then he said that he hoped that some of the comments that the author replied could be placed in the final text version.
He said, This work looks quite solid to me. The word solid is a very positive word. To say that a job is very solid means that he approves of this job very much.
After we responded to the review comments, he had already started to congratulate us, Congratualations to the authors. Although he didn't know who those authors were, he was already congratulating. He actually knew at that time that we would definitely win, because there was no negative score. There are many translation methods for solid. After the rebuttal, we actually finished the draft.
Looking at our screenshots on CVPR at that time, we were quite honored to win an oral article, and the acceptance rate of oral in CVPR was about 4%.
In fact, there are many stories in it, including your experimental verification, the improvement you made, including the iteration of those papers, including how we deal with those review comments. Some review comments are indeed real. Help us to improve, some reviewers' opinions are purely fault-finding, so the whole process has been close to more than a year.
At that time, after our thesis was drafted, a CVPR seminar invited us to do some presentations, saying that your work is more suitable for our workshop, so you invited me to do some presentations at the CVPR site. In fact, this year, CVPR held a meeting in New Orleans, the United States, but because of the epidemic, we couldn't go in China. This is the screenshot of his invitation to us at that time. After your paper is accepted, you may receive some such responses.
Then there is Tesla, the Tesla in the United States, their recruitment manager emailed me saying congratulations on your paper acceptance at CVPR 2022! Then introduce Tesla, say that we are a world-class AI robot and vision team, and then we are doing the most influential work in the world, and then what are we doing now, and then we say that we are doing Tesla’s automatic Driving systems, including humanoid robots. After introducing their own products, they said that our team would like to invite you to a closed-door meeting of CVPR 2022 , and then said that Tesla is hosting an exclusive, invite-only event , which means that only those who are invited can participate, An exclusive is a closed-door meeting, which means "close the door for a meeting", and then show some of their latest technologies, solutions, show some demos, and then he said you can meet with our engineers ... and also enjoy food and drinks in a fun setting. We'd love it if you could attend. Let me know if you are interested. This is Tesla's R&D team in the United States. The R&D team of robots and AI invited us to attend one of their closed-door meetings .
In fact, they didn’t invite us, they just sent me an email to invite me to attend, but it’s also a pity that the epidemic can’t get out, otherwise we can really go to the United States for the meeting. In fact, for foreign companies like Tesla, if you can recruit through these channels and get an offer, the annual salary they give you—from what I learned from my classmates—is basically around $150,000. Between 300,000 knives. This is the result of the last CVPR.
3. How to save yourself?
The above, in fact, I just want to express, you can spend a year, or even not need to spend a year, and you can do this thing. Now that the conditions are getting more and more convenient, you can do these things.
Now back to the present question: how to save yourself? I just shared my own experience. Now I want to systematically analyze what are worth referring to in this process and what may not be used. I will try my best to share with you the experiences that I think are useful. You can See if it is really useful and helpful to you.
1. How to systematically solve scientific research problems?
Then how to save yourself? Let's go back to this question: what should we do if we want to systematically solve scientific research problems? Going back to this picture again, it has so many modules, and each module is interlocking. If you solve all these modules by yourself, then you are really a master Yes, there must be students like this around you.
2. Why is it not easy to read and understand the paper?
Let’s look at each module separately, first of all, the thesis, and the link of reading the thesis, I summarized what problems there may be in it- English dyslexia , background knowledge , reading methods , and then the number of readings , including exchanges and discussions , These obstacles make it easy for us to get stuck in one place when reading papers.
3. Why is the idea so hard to imagine?
Then when it comes to the process of thinking about ideas, why is it difficult to think about ideas? In fact, there are many levels. You will have a difference in thinking, and you need to change your thinking. You have to sort out the ins and outs of this problem. This process is actually very time-consuming. And energy, many times you think you understand, but in fact you don’t understand, and then there is a logical gap, including some guidance on methodology, there is a methodology of thinking about how to find an idea.
Many times when you encounter a problem, you don’t know whether the idea is reliable or not. You need an experienced person to help you judge and let him tell you, according to his experience, whether the idea is reliable or not. Whether it can be done, whether to spend time and energy, or how much time and energy to invest in it. These all need some feedback and guidance.
4. Why is it not easy to write a thesis?
Is it easy to write? Writing is hard too! I still feel that writing is difficult. I have written so many papers. I still think it is not easy, because it has too much room for improvement, because everyone’s writing level is really very, very, very many. different levels, it is a skill that can be continuously improved. In fact, there are many, many problems that need to be solved. Things are difficult, and there is the conception of your idea, how do you conceive your own idea, how to lower the reading threshold for readers, how to tell stories, how to express this kind of writing, etc. All need to be done step by step.
5. Use all the resources around you to find guidance
Then I will give you an action guide. The core principle of this action guide is: use all available resources around you to find guidance . This should be the most practical way.
For me at the time, I also used all available resources around me to find those students who had already published papers, whether they were brothers or sisters, or even your juniors, as long as they are more experienced than you. You can do this, it's not shabby. If you want to submit a paper, you need to find some experienced people to help and guide you, and work together to write a paper, which is not shabby.
■ Learning by doing: learning by doing, learning by doing
How can scientific research capabilities be improved? In fact, it is very simple. Now I think it is not as mysterious as it is said, just do it. In the process of doing it, you will learn, and in the process of learning, you will continue to do it. It is like this to find out your own problems bit by bit, and then make up a lot of your own foundations. Then at this time You will have independent scientific research ability.
This is the core idea I want to point out, that is, you have to do it, not think about it . Even if many people are stuck on the first step and the second step, they still have to do it. They still have to be patient and read the papers. There are also a lot of experiments that should be done. Don't expect others to give you ideas about "ideas". Everyone lacks ideas, so how can they give you ideas?
■ self-study
Another point, if you feel that no one around you can help you, then help yourself. As I just said, what problems need to be solved here, and how should we go about intensive reading? How to read extensively? How to sort out the research context? Learn the basics? Thinking about the methodology of idea? You can do what I said, step by step, after you know how to do scientific research, you can do it, step by step, make up for what is missing, and solve any problems you encounter, if you are willing to spend time And energy, it can be done.
■ Find someone with experience
If you feel that it is time-consuming to rely solely on yourself, or you want to do some scientific research results that you think are temporarily beyond your imagination, for example, if you want to publish CVPR, then the more effective way at this time is to find some experienced people Come and take you, your seniors and seniors (generally seniors and seniors, most of them are seniors, and seniors speak less, mainly seniors, most of the examples around me are this situation, the ratio of male to female in the laboratory is relatively small).
For example, if you are a novice and you want to ask a question, why are they willing to take you? Why are you willing to help you? That's one question; the second question is, can he take you? Does he have the ability to bring you up and take you to publish top articles? This is the second question you have to ask; and the third question is, you have to clearly know that as a master student or an adult, although you may stay in school after finishing your undergraduate degree, but at this time Many people have the thinking of some students, thinking that there must be someone who is willing to help you unconditionally, but in fact, if you think about it, can you help others unconditionally? Except for your children, you may help, and your parents may help you. No one will take the initiative to help you unconditionally. This is a thinking change that everyone should have as an adult. A student’s thinking may say, oh, please help me, I am so difficult, so tired, and I am so sad... But you have to know clearly, if a brother and sister go to take you, how much he/she will have to pay Time and energy, wouldn't it be better for him/her to do it by himself? You must recognize this matter.
There is also this kind of situation, for example, if you buy food, drink, and do chores, go to build a good relationship with the senior brother little by little, and then there is a suitable time. The senior brother may have less time but some ideas want to be done. If it can be given to you, you can take over and do it. This kind of situation also exists, or some seniors may just enjoy guiding students more.
■ Find a reliable teacher with a professional structure
Later, you can find reliable teachers from some professional institutions.
Everyone can think about it for themselves and sort out their own needs. You need to know where your problems are, and then see if others can help you solve these problems? And can he do it?
You can find some senior PhDs, or teachers who have graduated and are working as algorithm engineers. They may also have some needs to publish papers and have experience. And what are they missing? They are short of money. It’s very simple. People want to make money. After they enter the society, they may marry their wives, buy a house, have children and raise children. All of these will cost money, so people are willing to use these experiences as a value exchange in this way. , he acts as an agent and asks you to publish papers, and then he can get a certain return, remuneration, then this is a matter of willingness, the rest is whether he has the ability to lead you to do this, and whether he himself has the ability Have you brought other students to publish papers? So what if he publishes the paper himself? Are you on the front line of scientific research? You have to think about these questions, can he really do it? Can he help you with these problems?
Now that there are so many resources on the Internet, you can look for them, find out how many reliable teachers there are, and then contact him and chat with him. Of course, you have to sort out your needs at this time. During this process, I think there needs to be a process of mutual understanding. Both parties need to have some understanding of each other before they can cooperate.