Looking for a topic? Breaking through bottlenecks? Teacher-student relationship? MSRA Joint Training Star Friends Guide You Through

This article is reprinted from the Microsoft Research Asia (MSRA) official account, thanks to MSRA for the invitation!

Pursuing a doctorate is a lonely and challenging journey. If there are senior brothers and sisters to guide you, they can definitely help you out of confusion and bravely make choices. On February 8, five alumni who graduated from the Microsoft Research Asia Joint Doctoral Training Program shared their personal insights during their doctoral studies and subsequent work online to answer questions and encourage everyone to stick to their dreams. This event was organized and hosted by Sun Lijun, Senior Academic Manager of Microsoft Research Asia, and Dou Anqi, the person in charge of the internship program.

We hope that the sharing of these senior brothers and sisters can encourage and inspire every one of you on the road to a doctorate, and let you enjoy this beautiful and challenging time~

Guest speakers:

• Fu Xiaoming, 2016 USTC-Microsoft Joint Training Doctor, currently Associate Researcher at University of Science and Technology of China
• Zhang Chi, 2017 Sun Yat-sen University-Microsoft Joint Training Doctor, currently DeepMotion Co-Founder, R&D Director
• Huang Danqing, 2019 Sun Yat-sen University-Microsoft Joint Training Doctor, currently a researcher at Microsoft Research Asia
• Li Bojie, 2019 USTC-Microsoft Joint Training Doctor, currently a senior research engineer at Huawei 2012 Lab
• Li Xiao, 2019 USTC-Microsoft Joint Training Doctor, currently a researcher at Microsoft Research Asia

Q1: As a doctoral student, how to handle the relationship with the mentor?

Li Bojie: First of all, students should let your mentor make multiple-choice questions instead of fill-in-the-blank questions. For example, if he asks you for an idea or solution to a problem, you can propose three or four even very stupid methods, and the mentor will choose a relatively good one from them. The second is to hope that everyone can maintain a relationship of both a teacher and a friend with the mentor, not a relationship of superiors and subordinates, do not hide, the mentor will not require you to invest all your time in him, you must be honest with him, say what you have.

Zhang Chi: I have personally seen some examples, some students are afraid to ask questions, which will slow down their progress compared to students who dare to ask questions. At first, asking questions may seem a bit stupid, but after asking, the level will get higher and higher. The factors that determine a person’s level are not only your current knowledge content, but more “learning speed X time = final total amount”. If you can increase your learning speed, that is, improve your internal skills, you will go further in the end.

Huang Danqing: During my PhD, I also had several mentors of different sizes. Some mentors may discuss some technical details with you, and some may just discuss the big direction and tell you a blueprint. My experience is that I have to guide my own mentor, know the style or strengths of each mentor. For example, some mentors are more senior, so you can broaden your horizons under his leadership. But at the same time, you have to try to lead him to the places you are interested in or puzzled, you need to communicate with him with questions. Now I also start to bring interns, standing from the perspective of a mentor, I hope that students can not only report results, but also have some thoughts from the results, and then we discuss along his thoughts. I also hope that interns treat me as a friend and disturb me more.

Q2: When was the research direction during the doctorate determined? Is it independently planned, or is it formed through discussion with the mentor?

Huang Danqing: I was very lucky at the time. As soon as I entered the group, the mentor said you follow this project, so my doctoral direction was set at the beginning. I think for those who are not mandatory by the mentor, you can see what most people in the group do, and then do some related ones, because there will be more resources in this way, more people can discuss together, and more ideas can collide together. Then you have to read more papers and see what the academic community is doing now.

Li Bojie: I think my paper was published smoothly because it just “grasped the pulse of the times” and did what everyone in this field was interested in, which is to use programmable hardware to accelerate the network or system. If it is only from the perspective of publishing papers, it is still necessary to grasp a direction that the research community is interested in, and at the same time combine the advantages of your own company and research institute, it will be easier to succeed.

Q3: During your PhD, how did you improve your coding ability and English ability?

Li Bojie: I think participating in company products is quite helpful. Not only can you go deep in the research direction from the vertical, but you can also participate in the company’s projects horizontally, and theory and practice can be well combined.

Participating in a project can be considered as drawing a line. If you draw 4 lines into a tic-tac-toe, it becomes a square, and this square can net a fish. If I draw more lines to form a net, then all fish can be netted.

In fact, we can improve our coding ability by doing various projects, whether they are internal projects of the company or external open source projects.

Li Xiao: As for English ability, listening, speaking, reading and writing are actually not the same. For listening and speaking, on the one hand, you have to speak when you give a report, and more often it is during the meeting process and even possibly outside of these parts, such as discussing issues offline with others, chatting based on other people’s posters, and even catching some big shots and chatting with them. For example, going to CVPR, and big shots, peer researchers can communicate, not only is a very valuable opportunity to practice English, but also a chance to chat and gossip.

As for reading and writing, especially writing, you may have to write more, at least I am like this. I was very miserable when I first started writing, but after writing more, it will be better.

Zhang Chi: I agree with Xiao Di (Li Xiao)’s statement. English ability is divided into two aspects, one is doing academics itself, and the other is chatting with others. I have seen a research before that says that everyone’s ability to master a foreign language is very related to his own field. It may be difficult to change to a relatively unfamiliar field and let us chat with others about football, rugby, foreign stars. We should try to create an immersive environment for ourselves. Once we enter this environment, our abilities will gradually improve.

Q4: During the process of pursuing a PhD, there will always be many difficulties and bottlenecks, such as lack of progress in work, inability to publish papers, how should such problems be solved?

Fu Xiaoming: When I was pursuing my doctorate, I also encountered similar problems, and I managed to withstand them relatively speaking. One simple way I dealt with it at the time was to go to my mentor when I had nothing to do, and chat with him for 10 to 20 minutes every day.

At MSRA, many times there may not be some fixed objects to vent to, so I can only bother my mentor. Only the mentor can give me some professional advice. Now from my own experience of mentoring students, if the teacher is more active and talks to you more, you will be less confused. Try to communicate as much as possible, which may help you to better release the confusion in your heart.

Li Bojie: I quite agree with Professor Fu’s statement. Whether I am a small mentor in the company or I am looking for my own mentor in school, mentors will not find it annoying, but they will find that students report too little. Sometimes when a mentor assigns a task to a student, it disappears like a stone sinking into the sea. The tutor has to take the initiative to ask, and then the task will have a result. Timely feedback on tasks and closed-loop awareness are very important, and everyone needs to improve their initiative.

Huang Danqing: I have also encountered the problem of no progress in work. My experience and the advice my mentor gave me before is that whether you persist or terminate, you have to have an analysis: why can’t you do it, where are the challenges, etc. You have to give yourself a reason, you can’t say I don’t want to do it, or I can’t do it after half a year, so I won’t do it.

Li Bojie: Now I may not have much pressure to publish papers, and I feel that being rejected is a good thing. There will always be many poorly written parts in the papers I write. I always write in a hurry. The reviewers give a lot of opinions. Those comments can indeed improve the quality of the paper. After you make changes according to the requirements, you will have the feeling of “no regrets after sending it out”.

Zhang Chi: In fact, my first paper in life was also rejected. I was very depressed at the time. The biggest gain later was to form a set of psychological adjustment mechanisms through this process, trying to eliminate the knot caused by the failure to publish the paper. It will definitely affect the mood, but we still need to face it positively.

As Bojie said, after being rejected, if you sort it out immediately and change it into a new version, you will feel better. The reason for the bad mood may be because your work is not good enough and is not recognized. But if you come up with something new, it may be recognized, which is a positive thing.

Q5: The overall industry is now “inwardly rolled” quite seriously, how to balance research and life?

Huang Danqing: I think it is necessary to ensure normal rest and exercise, because for research ideas, sitting and reading papers all day is not very useful, it is better to go for a run to change your thinking. Some people may say that I do experiments and have to work overtime until 12 o’clock for several days, but I think this is just a short-term imbalance. If you have to do experiments so late for half a year, there must be a problem with your coding, or the idea is not mature in the first place, you have to adjust yourself. You still have to spend time resting and dating.

Li Bojie: I think in fact most people have very little effective working time. Many people sit for 12 hours, but the effective working time may be less than 6 hours, especially under the current conditions, the computer can browse various websites, and the mobile phone can play everywhere. I think maintaining a higher work efficiency is more important than maintaining a longer working time.

In addition, I think regular thinking or review is a more important thing, such as taking out two hours every week to think about the past things, at this time there may be many new ideas, which is more efficient than I keep doing things, and it will also have a better sense of direction for future work.

Zhang Chi: Some people may think that the longer they stay, the more output the company will have, and the more money they will earn. But at least in the AI industry, or industries that rely on intelligence to produce, this is definitely not the case. Like programmers, don’t write code when you’re tired, otherwise you’ll have to debug the bugs you write later.

The same is true for scientific research. Blind pursuit of 996 will not bring more scientific research inspiration. Students may feel that 996 is not very tired, there is only one thing to do, but after really working, there are more things to be busy, combining work and rest can improve work efficiency.

Li Xiao: Now my workstation is next to Teacher Huo Qiang from the Speech Group. Teacher Huo, including my mentor Guo Baining when I was reading the joint training doctorate, both said a sentence I think is very good, “Doing scientific research is more like a marathon rather than a sprint”. Short-term 996 may not be the most effective method for scientific research. I think scientific research often requires you to communicate with others and share your views with others. Some ideas may even come out when you are chatting with others.

On the other hand, from the perspective of methodology, for fields like deep learning, we often need to run various experiments and draw conclusions based on experimental results. I think a better method is to do as much automated script processing as possible in the early stage, which is also what my mentor shared with me. You don’t have to 996, but you should try to make all the computing resources you have 996.

Q6: Under the current situation, how to choose between academia and industry after graduation?

Fu Xiaoming: First of all, follow your heart and ask yourself what you want to do. I know more about academia. Although the “inward roll” of academia is now very serious and the pressure is also great, as long as you have academic enthusiasm, you can still come to academia.

Huang Danqing: I was jointly trained at Microsoft and also work at Microsoft, so I will talk about the process of my choice. I don’t think I am particularly good in academics, so academia may not be suitable for me. For outside companies, I heard many people say that 996 or the pressure is relatively high, so I don’t consider it much.

At MSRA, I have been with Professor Lin Qinyou for a long time. He is a trustworthy teacher who can bring me more work ideas, so I think it is very important to follow a good boss.

In fact, I think MSRA is a place that balances industry and academia quite well. The academic freedom here is quite large. Although everyone emphasizes the importance of product transformation now, I think under such a background, you also have many scientific research projects that can be done. I am quite satisfied with this state.

Li Bojie: I agree with what Dr. Shen Xiangyang said. Your research career is composed of a series of projects. If you find good projects and good bosses on a large platform, there won’t be too many problems.

If you are not completely sure where to go now, it is best to interview several companies, and you can also go to several places in the school to see, and understand whether this place is the same as you imagined, because many people enter After that, I found that it was very different from what I imagined, and it was very painful.

Zhang Chi: In academia, you have to be more responsible for yourself. For example, if you are a teacher, you have a research group and research topics, and you have to be responsible for everything. If you are in a company, like us, we already have a big goal, a team goal, which is a team working towards this goal.

Assuming we turn the question of choosing industry and academia into a target function, in addition to industry/academia, your own personality is the biggest variable in this target function. You must figure out what you like.

Li Xiao: If everyone is not clear yet, you can actually go through it in your mind, list all the offers you have received, or all the places you want to go, measure it from several dimensions, such as salary, work freedom, and ease of work, etc., and compare it with the ideal job in your mind to get a final conclusion.