Foreword : I learned about this book at an academic sharing lecture given by Mr. Zhang during the lockdown period last year. I thought the content of this book was commonplace, so I didn't pay much attention to it. Later, I borrowed a copy of "Smart Agents That Love to Make Mistakes" written by Teacher Zhang from my junior student. I found it very interesting to read, so I found this book to read. After reading it, I realized that my first impression of it was biased. This book covers scientific research thinking, scientific research methods, scientific research topic selection, scientific research quality, scientific research ability, mentality adjustment, paper writing, submission rules and other aspects on the road of scientific research that we may feel confused or overwhelmed by choices. The vast number of master and doctoral students have provided very valuable and reference "tips" for postgraduate study. It can be seen that the author is very dedicated to imparting his accumulated years of scientific research experience, which has benefited people a lot. It is strongly recommended that students who have not yet started a research career or scientific research whites read it as an introductory book on scientific research. The following are some notes I took during the reading process, and I would like to share with you~
sequence
-
Postgraduate study is mainly to broaden horizons, exercise abilities, and prepare for the future. In this sense, what topics to do during graduate school is not the most important. It is necessary to cultivate one's ability to "be able to achieve results on any topic" and " be extraordinary in the ordinary ".
-
One day many years ago, I walked through a small street outside Princeton University, and was attracted by a small store on the street, not by its product advertisements, but by a plaque. The plaque reads "Einstein Exhibition Hall". There is actually an Einstein exhibition hall hidden in the small shop? ! I couldn't resist going in to see what happened. I found that there is really a small room with the same plaque hanging in the back hall of the store. There are not many furnishings in the room, but a quote from Einstein on a column "Imagination is more important than knowledge" left me deep impression. The words of the master are loud and clear, pointing out the key driving force of many creations and inventions.
-
If there is no profound knowledge as the foundation, it is impossible to make important innovations by relying on the imagination of "unconstrained style and unconstrained style".
01 Shishang, starting from Erdos
-
Graduate students must not be overconfident, thinking that they can spend less time on scientific research after playing games or after finishing other chores, so that they can outperform those who may be smarter and work harder.
-
Thinking every day, dreaming at night, this method can be used for reference~ In dream learning, the most basic method is to close your eyes before going to bed and review the relevant research you have done that day, sort out your thoughts, and then hint that you can dream about related research at night. content. In dream learning, the most basic method is to close your eyes before going to bed and review the relevant research you have done that day, sort out your thoughts, and then suggest that you can dream about the relevant content at night.
02 Time Quotient and Basic Misconceptions
- In addition to the strong pertinence of the research direction, compared with laying a solid foundation from books, reading papers may have an additional benefit, that is, it is not easy to be bound by the rules and regulations in textbooks. There is a saying that goes well, the more knowledge you have, the harder it is to think. **Human innovation often starts with making mistakes, which can even be equated with making mistakes. **If we abide by or believe too much in the norms and guidelines in textbooks, in fact, we tend to limit our imagination or imaginative space, and dare not think critically. Therefore, we need to avoid basic misunderstandings more or less.
03 My imagination takes flight
- The sources of imagination are different, and some are even quite different, but in summary, there are two things in common, which may be an important source of imagination: one is neoteny, and the other is a wide range of hobbies . "Neoteny" (Neoteny) is also called "neoteny continuation", which is an important concept in sociobiology, that is, the process of slowing down the maturity. "Guardianship" and nurturing, until the growth process of independent livelihood or self-reliance, neoteny in a broad sense even includes pregnancy. The love of learning and curiosity about new things and knowledge can all be regarded as an external manifestation of juvenile persistence. At the same time, neoteny also delays the growth of mental age. It can help people avoid some inherent rules and regulations as much as possible, constantly step out of their comfort zone, and stimulate interesting and beneficial imagination. To put it simply, if you want to do a good job in research, it is best to maintain a childlike innocence that is always curious about the world.
04 The tyranny and dopamine of Hunan people
- Eat hard, endure boredom, domineeringly, not afraid of death. (Perseverance, patience, perseverance, determination.) Scientific research and innovation also require such qualities. "Eat bitterness" means not pursuing too much or caring about the comfort of the environment. // Be patient, it was not smooth sailing at the beginning, although this idea is feasible intuitively, but computer science is an experimental science, and the actual data is noisy. // The second meaning of being overbearing is that once Hunan people become overbearing, they like to talk to the end, do things well, and will not back down halfway. This is precisely the quality that is especially needed in scientific research. Some students may shrink back when they encounter a little difficulty, so that they frequently change directions during graduate school, which leads to a serious reduction in the time spent on actually delving into a scientific problem. On the contrary, without the spirit of challenging the limit, it is not easy to achieve good results in scientific research. For example, after doing scientific research for a long time, a comfort zone has been formed, and some people are unwilling to open up new directions. A similar lack of spirit of challenging the limit will also appear in young students who are afraid of failure and worry that failure to produce results will affect their development.
05 Joking about finding a teacher
-
Matsushita asked the boy, saying that the master was going to gather medicine. Only in this mountain, the clouds are deep.
-
After all, psychology has also said that most people who are usually very docile and low-key will be terrible and uncontrollable when they get angry.
08 "Bubblegum" and talk shows
-
"I'll create opportunities for you, it's up to you whether you can seize them or not."
-
"Bubblegum" production skills: But if you write all the content on the PPT, and then read the PPT without changing the tone, it is obviously impossible to attract the audience like a talk show, and it is impossible for the audience to have expectations for the unknown content in the report .
09 Seminars and Comprehensive Research Ability
-
The rope of learning is stretched so tight for a long time. Over time, students' curiosity and interest in new things will be obliterated by this high-intensity learning. Therefore, the ability to ask questions in class will also be lost.
-
Therefore, in the discussion class, encourage graduate students to push the formulas in the report papers more carefully, which is not only beneficial to the students’ understanding of the problem and improving their ability to react on the spot, but also helps the students listening to the audience to better see the innovation of the paper Point and identify potential research directions.
11 Selecting the topic and title of the thesis
-
Thesis writing is crucial to improving the logical thinking ability of postgraduates in scientific research, such as the logic of continuity and rigor, cultivating perfectionism and teamwork, etc.
-
Try to be cautious when selecting topics, be targeted, and explore more different research directions.
-
From my experience point of view, I still recommend writing the title as clearly as possible to ensure that it can be assigned to a knowledgeable reviewer. Because even if it is rejected, more professional opinions can be obtained, which is convenient for the improvement and perfection of the next step. After all, the purpose of writing a paper is more to disseminate your research results better, and to further improve your scientific research capabilities, not just to publish a paper. This should be the original intention of researchers to do research.
13 Catchy Openers
-
In April 2013, on the Gold Coast of Australia, under the blue sky, Professor Yang Qiang of the Hong Kong University of Science and Technology swam fast and rhythmically in the outdoor swimming pool of the hotel. When he stopped, I went over to say hello to him, and asked about the situation of my master student, Tan Ben, who was studying for a Ph.D. with him. (I like the beginning of this~)
-
In addition, the statement here also needs to be linked to the subject of the submitted conference or journal, so as not to make the reviewers feel that the submitted article is not within the scope of the permitted submission. For example, if it is a conference in the field of computer vision, the author can point out in the language expression that biometric authentication mainly relies on processing technology related to video or images.
-
It is best to summarize in an inductive way, dividing the problem-solving strategy into several different ideas. For example, the strategy of gait recognition can be decomposed into three categories: modeling based on time-series videos, summarizing single image modules from video sequences, and models based on structural elements such as matchstick structures. Then for each line of thought or each category, point out the representative literature among them.
-
In the overview, what also needs to be written is to point out the problems existing in the different category methods in order to lead to the new methods to be proposed in the paper. These problems can exist in a single paper, or they can be common problems in a class of methods, but they must be closely related to the content of the paper being studied or the new method proposed. It should be noted that since it is scientific research, it must be clear that there is no ultimate algorithm, and each method always has shortcomings or problems. Therefore, if some problems in the research direction are not intended to be solved in this paper, there is no need to write them out and put them in the main text, and only analyze the key issues related to the paper.
-
For example, it can be considered from the innovation of design ideas, theoretical proof, improvement of algorithm optimization efficiency, significant reduction of algorithm calculation complexity, effective improvement of model structure, and reasonable adjustment of objective functions. But it should be noted that the contributions or innovations mentioned should be used to solve the problems in the previous literature. In this way, a logical closed loop can be formed in the introduction part of the article. In addition, if the experimental effect is good, is it considered a contribution point? Some reviewers think that this is not a substantive contribution point. Therefore, when the innovation point is not enough, it is not too late for the author to consider using the full verification of the experiment as a contribution point.
-
In short, in the introduction or introduction part, the content should not be too much, so as not to hinder the reviewers and readers from quickly understanding the innovation points; the first part of the review should not be too broad, choose important work that needs to be solved; innovation points It needs to be condensed to show the truly innovative points. It is best to be able to understand at a glance and express them in points and lines.
15 Innovation——From Huntington's Disease
-
In the direction of artificial intelligence, there are mainly two types of points that can form innovations, theory and algorithm. Then the author should write clearly in this section, such as whether the theory and algorithm have solved the problems mentioned before, and how to solve them.
-
Some students like to use longer strings when defining mathematical symbols, such as "LongString", "Variable", "MathSymbol" and so on to define a variable. Although there is nothing wrong with writing in this way, it is easy to make people feel that the author's mathematical training or literacy needs to be further improved. Defining symbols should follow the principle of "simplify as much as possible without affecting understanding". For example, the above three symbols can be simplified into ls, var, and ms.
-
If your theoretical level has not reached a certain level, don't be overconfident and force yourself to do some "superior" proofs, especially for articles in top journals or conferences. Because you may meet scholars with much better foundation than you, they can see your problem at a glance, and may magnify the deficiency of this problem, and use it as a reason for rejecting the manuscript.
16 Complete experiments
-
Regardless of the cost, in the experimental part, the author should try to explain the computational complexity of the proposed method or the computational time shown in the experiment to help readers intuitively understand the actual performance of the new method.
-
The realization of the process of subjective evaluation can be completed through online (paid) anonymous questionnaire filling, such as using the Amazon Mechanical Turk platform, or you can build a link to an online website yourself. Through this method of analysis, the indicators of subjective evaluation can be as objective as possible.
-
The purpose of the experiment is not only to verify the superiority of the algorithm, but also requires the author to be able to analyze some meaningful and valuable knowledge from the experimental results. On the one hand, this knowledge can help the paper form a logical self-consistency. For example, whether the deficiencies of other methods mentioned in the introduction part of the paper, and the innovative points of the method proposed in this paper are fully reflected by experiments. If not, then you need to go back to the previous chapters, readjust the tone of the paper, remove the content that may exaggerate your method, and keep the logic without loopholes. Such an adjustment cannot be completed once, and often requires multiple iterations.
-
Because for most applied research, the Terminator's algorithm does not exist, there will always be some problems. As long as the analysis of the problem is reasonable, the author does not have to worry about the reviewer rejecting the manuscript. Of course, don't belittle your own research results too much, that won't help the acceptance of the paper.
17 Anticlimactic conclusions
-
As the saying goes, "Say important things three times". You may wish to do the same in the paper, lest the reviewer forget what the important innovation of the paper is. At the same time, some constructive thinking should also be provided, including: (1) where the proposed method has potential applications. Taking the application of research on facial micro-expressions as an example, the paper can emphasize that this research can be used to evaluate the true emotions of humans, and has potential applications in bank credit evaluation. (2) What are the research directions that can be carried out on this basis in the next step. For example, real-time micro-expression detection and recognition, or higher-precision detection and technology. Great care should also be taken in language organization to avoid obvious spelling and grammatical errors. (3) In order to prevent the reviewers from asking questions, they can actively raise the deficiencies in the paper. For example, there may be small limitations in your own method. At this point, it can be stated explicitly and pointed out that it will be studied in future work.
-
When presenting your own results in the conclusion, there is a balance, neither too boastful, nothing out of nothing, but also not too low-key. This requires tutors and students to control this degree together.
-
There is also a reference to a book, which also needs to be carefully checked by the publishing house. Because it has direct sales agencies and chain agencies, such as Springer Publishing House, in addition to its headquarters in Berlin and Heidelberg, it has branches in many places such as New York, London, and Paris. Therefore, when citing, it is not enough to write Springer alone, and it is necessary to write the name of the specific chain publishing house.
18 Perfectionist Typography
-
In the primitive society, there were many natural enemies for human beings. If you want to save your life, the most effective way is to be able to quickly identify the outlines of natural enemies, such as tigers and lions. If you look at the details of the natural enemies first, such as the texture of the tiger, you may have already lost your life. Relying on the ability to quickly identify a wide range, reviewers can also quickly understand the level of papers. In other words, without looking at the details first, you can directly reduce the paper to only the framework structure, and you can judge the general level of the paper by looking at the layout of the paper.
-
In addition, if the paper submission system allows it, you can consider giving a version with 1.5 times the line spacing when submitting the review paper, which is convenient for review and reading, and can also leave a good impression.
-
When drawing a model structure diagram or schematic diagram, the color configuration should be as beautiful as possible, and no strange color matching should be used. It is also not recommended to use colors that are not obvious when printed by some printers, such as green and yellow, because some reviewers prefer to print out to read papers.
-
When drawing the experimental method comparison diagram, if there are many curves representing different methods, you can consider changing the line type (such as ——, -.-., —) and adding marker points (such as *, +, five-pointed star, diamond, etc.) to help distinguish
-
Avoid duplication and prevent wasted pages.
-
Mark key results explicitly. For example, the most effective results in the table should be displayed in bold or italics, and the second-best results should be underlined or wavy above them to distinguish them from other results, so that readers can see the differences at a glance without wasting time to find out by themselves.
-
It is recommended that authors include the cross-reference function when writing an article. For example, after using this function in LaTeX, the page number of the corresponding reference position in the text will be provided after each cited paper in the reference. In this way, it can be convenient for others, and it is convenient for readers or reviewers to quickly know which citation of the paper is. Second, it can also ensure that all references are cited in the text, rather than "filling in" random additions.
-
In order to maintain the flow of reading, the position of the figure should try to stay on the same page as the position where the figure is first referenced.
20 Win-win cooperation
-
There are two well-known registration-free conferences in the field of artificial intelligence. One is the "China Machine Learning and Application Symposium" (MLA for short), initiated by Academician Lu Ruqian of the Chinese Academy of Sciences at Fudan University in 2002 and promoted by Professor Zhou Zhihua of Nanjing University. ), the meeting time is the first or second weekend of November every year, and mainly invites well-known scholars in the field of machine learning at home and abroad to give reports, as well as experience sharing on top conferences. At present, it can attract 1500-2000 people to attend the conference every year. The other is the "Vision and Learning Young Scholars Symposium" (VALSE for short), which aims to provide a platform for in-depth academic exchanges for Chinese scholars in the fields of computer vision, image processing, pattern recognition and machine learning. The number of participants in previous years has reached 5000. Because it is more practical, there are many sponsors of the conference, so not only the registration fee is waived, but also free lunch and dinner.
-
Due to the limited distance and unsmooth face-to-face communication, remote international cooperation may be more prone to disagreements and even lead to interruption of cooperation. Especially students, more attention should be paid to grasp the propriety. The interruption of some cooperation is often because students are overly curious or confident, and they believe too much in the accuracy of their own judgments, and tend to deny the ideas or viewpoints put forward by the other party during the cooperation. Going back and forth will delay the progress of the cooperation project, which will eventually lead to a loss of trust between the two parties. Therefore, this kind of cooperation requires both parties to make appropriate compromises in order to have the possibility of long-term cooperation.
-
It should be avoided that after declaring the project, they do their own things, so that the overall contribution is only equal to the sum of the parts, and no major results can be condensed.
-
For students who are capable of being independent, it is enough to provide an opportunity. Intervention is to limit students' performance with their already limited knowledge.
-
Most of the students' scientific research ability is between the two. What the tutor should do is to provide various opportunities, such as opportunities to participate in conferences, opportunities to find opportunities for domestic and international cooperation, and opportunities to explore research directions suitable for their scientific research capabilities. and so on.
-
For students, trusting their mentors is probably the most important thing. After all, there is always a reason why a mentor can become a mentor, and it will not come out of thin air. The experience of a mentor can often make scientific research more effective.
-
Only when the tutor and the students understand each other and perform their own duties like the "assembly line" emphasized before, can it be possible to improve the common time quotient and achieve a win-win situation for teachers and students.
-
Cooperation is not necessarily better than working alone, because not everyone and all directions are suitable to produce results through cooperation. Therefore, you don't have to think that your research will be at a disadvantage just because others have cooperated and you have not. Sometimes important results or flashes of inspiration are made by one person.
-
**Beasts always travel alone, but cattle and sheep are in herds. **If you feel that you are more suitable to be a lone wolf in scientific research in all aspects, that's okay. As long as you can identify and stick to one direction without wavering, you will be able to make a name for yourself sooner or later.
21 Indomitable scientific research, local application
-
It is not easy for the school to build a long-term and stable scientific research team. It is really "a solid camp with flowing soldiers".
-
Patent is also an important part of writing "application", which protects the intellectual property rights of scientific research results to a certain extent.
22 After scientific research
- If I use an image metaphor, I think Houlang should have the ability to resist blows like Hanamichi Sakuragi in the Japanese anime "Slam Dunk". When coach Anxi criticized him for his low shooting percentage, he still believed that he was a genius; after being rejected 50 times by girls in three years of junior high school, he still believed in love.
23 Unpopular and popular research
- A man was very hungry, so he paid a penny to buy a sesame seed cake. He was not full after eating it, so he bought another one, but he was still not full after eating it, so he bought another one... and so on, until he finished eating the seventh sesame seed cake , I feel full. He patted his bulging stomach and said to himself: If I had known earlier, I would not have eaten the first six sesame seed cakes. (The first six biscuits can be regarded as accumulation. Without the first six biscuits, eating the seventh biscuit will not feel full. This is a process from quantitative change to qualitative change.)
24 Psychological hints and challenging the limit
-
The book "How to Suggest Yourself" written by Japanese writer Dahu Hui (this book is now translated as "Self-Suggestion".)
-
When you can do active research, don't get involved in other things.
-
Some students are more sensitive to the environment and cannot stand a relatively noisy environment. In a sense, this is caused by insufficient anti-interference ability, and it may also be related to the overemphasis on quiet in the learning environment since childhood. This may as well give yourself some psychological hints, imagine that the future working environment will be like this, but you can still make good results. (It resonates quite well, so I like the study room.)
-
Trying some delicacies may promote the positive activity of intestinal flora, and may help relieve the physical and mental stress caused by scientific research. If you are worried that this method will easily lead to weight gain, you can also increase exercise appropriately. After all, the dopamine release that results from exercise can further make people happy.
25 Curiosity and details determine success or failure
-
If you want to gain a sense of accomplishment in your dissertation, you might as well train yourself to be a perfectionist in dissertation writing, and be critical of your own research results, in order to gain a better and more effective sense of accomplishment.
-
The credibility and repeatability of the experiment is guaranteed by scientific design of the experiment, the generalization of the model is tested on different data sets, and the rules and interpretability are discovered by analyzing the influence of each parameter. At the same time, it is also necessary to consider the limitations of the proposed model and find potential breakthroughs for further improving the performance of the model. The more elaborate the experimentation, the better the method will stand the test of time.
-
In general, if you can learn to turn the ultimate pursuit of perfect details into a subconscious reaction in your graduate career, you will be more likely to turn your curiosity into a stable motivation, and you will gain more in your graduate work and future work and scientific research. Bigger and more results.
26 Teachers and Students - Xuanwu Lake of Rabbits
-
In the marathon, when the runners are worried that they can't run, they will find someone who is good at long-distance running to run with them. That person is commonly known as "rabbit". During the marathon, many rabbits with different paces will be arranged, and the runners can run according to their own pace ability.
-
For runners who are just learning to run, it may be more appropriate to start running with a more energy-efficient natural run that is closest to walking—that is, a high stride frequency and a small stride as wide as walking, such as 180 steps per minute cadence. To gain speed, you can increase your stride frequency first, then your stride length, and then use your body to lean forward to use the force of gravity to accelerate.
29 Nobel Prizes for Slow Nurturing: Free and Useless Souls
-
Only when more and more people do not judge the value of life by the amount of wealth, can more people be willing to slow down and devote themselves to innovations that are beneficial to human progress.
-
Li Hongbing, an alumnus of the Department of Journalism, believes: " The so-called 'freedom' means that thoughts and academics, and even the concept of life can wander freely in the boundless space and time; 'uselessness' means the intentional alienation of the real utilitarianism around us. "
30 Poetry and the Distance
-
The siphon effect, also known as the siphon phenomenon, physically means that due to the gravitational force and potential energy difference between liquid molecules, the liquid will flow from the side with high pressure to the side with low pressure. Because the water surface of the nozzle is subjected to different atmospheric pressures, the water will flow from the side with high pressure to the side with low pressure until the atmospheric pressure on both sides is equal, the water surface in the container becomes the same height, and the water will stop flowing. The water in the container can be drawn out quickly by using the siphon phenomenon.
-
"As long as there is scientific research in mind, there are laboratories everywhere."
I hope it’s not too late to read this book, let’s work together!