Shulou(Shulou.com)06/02 Report--

2019-09-05 15:34:15

Guest of conversation

Professor Zhu Songchun (Song-Chun Zhu)

Malt Award, Helmholtz Award winner, Professor UCLA, IEEE Fellow, founder of Dark object Intelligent Technology

Dr Shen Xiangyang (Harry Shum)

Global Executive Vice President of Microsoft, Foreign Academician of the National Academy of Engineering, Foreign Academician of the Royal Academy of Engineering, ACM Fellow, IEEE Fellow

Moderator: Dr. Gang Hua

Dean and Chief Scientist of Bee artificial Intelligence Research Institute, IEEE Fellow, IAPR Fellow,ACM Outstanding Scientist

Host: as we all know, Professor Zhu Songchun and Dr. Harry are good friends for more than 20 years. They have made outstanding contributions in the field of computer vision and artificial intelligence. They are leaders in the academic and industry, and leaders of overseas Chinese scholars. Around 2000, they gathered and trained a large number of outstanding young students at Microsoft Research Asia and the non-profit organization Lianhuashan Research Institute in Hubei Province. now these students have made remarkable achievements and become the pillars of the academic and industry.

Both teachers like Jin Yong's novels very much, Harry especially likes Linghu Chong in The Smiling、Proud Wanderer, and Song Chun likes Xiao Feng in The Demi-Gods & Semi-Devils best. At that time, they agreed to have a "Huashan on the Sword" 18 years later. Therefore, I think today's dialogue is actually a discussion between "Ling Hu Xiang Yang" and "Xiao Songchun".

The topic of this dialogue is: road Choice in the Age of artificial Intelligence

First, talking about the development trend of artificial intelligence: the golden era of AI in the industry and academic circles.

Host: my first question is that both teachers have been working in the field of artificial intelligence for many years. What do you think is the development trend of artificial intelligence in academia and industry in the next 18 years?

Shen Xiangyang: thank you very much for giving us such an opportunity to learn from each other. It really feels like a master. My computer vision "Huashan on the Sword" with Songchun began almost in 2000. In fact, it was originally agreed to fight in 2018, and Songchun also said at that time that not only the two of us had to fight, but also that each of us had to bring 18 disciples to fight together (laughter).

First of all, I think we people are certainly lucky with the development of artificial intelligence today. When I was a graduate student, I focused on areas such as computer vision and robotics, but in fact we didn't have many good job opportunities when we graduated in the 1990s. Especially at that time, the development of computer vision and natural language processing was relatively slow, and there was not much application of landing scenes, but it developed very fast in recent years, which can be said to be changing with each passing day.

Personally, I think that in the next decade or so:

Artificial intelligence in industry, in terms of perception may usher in the golden decade, there are many systems can do, and can land a lot of application scenarios, in industry, whether employment or entrepreneurship, there will be a lot of good opportunities.

Artificial intelligence in academia, Songchun just described the development trend and prospect of artificial intelligence from six aspects in the conference report. Personally, I think the most exciting direction is the combination of brain neuroscience and artificial intelligence.

Not only does artificial intelligence have a golden decade in industry, but in the next 25 years, it will also be the golden age of artificial intelligence in the field of scientific research.

Zhu Songchun: I am very grateful to Professor Liu Yunhao and the conference for providing such a platform for dialogue. I've had this conversation with Harry for a long time, and today we can finally sit together and talk. In particular, everyone is more concerned about this topic and would like to listen to it.

Harry and I are both "post-65s". Some people have said that the generation born in China in the 1960s and 1970s was relatively lucky. At that time, the social atmosphere was very positive, people advocated science and technology, and many people had a sense of social responsibility and mission. But the problem is that when we went to college, we wanted to study computer vision and artificial intelligence. At that time, there were basically no professors in China who could guide us, and there was no access to information, so we chose to go abroad for further study. In the United States, we are guided by masters in our studies, but there are not many successful Chinese people in front of us who can provide reference on how to plan our careers. We all grope our way in the dark. Later, Harry went to industry and became a well-deserved industry leader, while I stayed in academia to continue to think about some of the problems that bothered me. As Harry said just now, it is very difficult for us to find a good job when we graduate in the 1990s. The two of us are walking "at night", and there is no one in front of us, so we are still quite afraid. Therefore, at that time, we often communicated on the phone and talked to each other about some career choices. It's like two people walking in different areas in the dark, shining a flashlight into the sky to see where each other has gone.

My opinions on how artificial intelligence will develop in the future are as follows:

In the development of academia, I have just made a report at the conference entitled "artificial Intelligence: towards an era of Unification". In other words, several fields of AI break away from the expression and calculation mechanism of mathematical logic, and after more than 20 years of exploration, we have found the new mathematical basis of probability and statistical modeling and random calculation, and on this basis, we began to integrate and move towards a unified pattern. I have preliminarily summed up these six major changes and trends in my report.

In the development of industry, I think the technological revolution of artificial intelligence is very different from the previous three technological revolutions. For example, the Internet and information technology revolution from the late 1990s to the early 2000s (now known as the third Industrial Revolution) is actually a relatively simple and mature application technology without much uncertainty. most companies are just doing business model innovation. Artificial intelligence is a very complex problem, the water is very deep, its application scenarios and tasks are often difficult to isolate and define, face recognition is a special case! Beware of a so-called AI-complete problem here: you only want to solve problem A, but it turns out that you need to solve problem B, otherwise you can't solve A, and then, in order to solve B, you have to solve C until you solve all the problems, which is general artificial intelligence. Ten years ago, I was saying a slogan that sounds not so scientific and a little funny: If you cannot solve a simple problem, you may have to solve a complex one! Popularly speaking, you need to solve a system of equations with 1000 variables, and it is often impossible to solve 3-5 variables alone. Now I can see that many friends in the industrial sector have not yet tasted this pain, and newborn calves are not afraid of tigers (laughter).

From my own experience, I proposed image interpretation (image parsing) and video interpretation around 2000, bringing all visual problems into a unified framework to find the best solution. Later, it was found that photolysis of visual problems can not be done well, and a lot of cognitive reasoning is needed (that is, intelligent dark matter I mentioned). At the same time, in order to improve the efficiency of learning, small data, big task mode, we must integrate language dialogue, robots and other fields.

Second, on the differences between industry and academia: both internal and external studies

Host: in fact, Professor Zhu Songchun and Dr. Shen Xiangyang both started their research careers in the field of computer vision, and then embarked on the road of academia and industry respectively. But both of them straddle two worlds. For example, Songchun is now out to start a business and founded DMAI, while Harry works as a part-time or honorary professor in many universities. So what are the differences between artificial intelligence in industry and academia? Can these two experiences help each other?

Shen Xiangyang: in fact, our Microsoft Research Institute has done a lot of academic research, and I have also brought a lot of graduate students. I would like to mention in particular the impact of the birth of the Internet on China's scientific research. After the emergence of the Internet, it has brought great changes to mankind, including a great impact on scientific research methods. For example, the Deployment-driven Research I advocated in Microsoft Research is to make full use of the Internet and user feedback. I personally think that the Internet can rank among the top three of the greatest scientific and technological innovations in human history.

However, for the industry, we can not only stay in the scientific research stage, but also have the industry to land. I often say that it is also good for you to publish papers in industry, but it is not the most important thing, but whether your scientific research direction and topics are leading and forward-looking. Moreover, when doing research in industry, the publication of a paper often means the beginning of a project, not the end. Of course, we also attach great importance to scientific research and academic papers. Over the years, Microsoft Global Research has five Turing Award winners, and of course they also write academic papers. If you do do great things, you will be truly respected. What I want to say is, don't write a paper simply for the sake of writing an academic paper.

Zhu Songchun: the relationship between academia and industry is compared with martial arts novels. To do research in the university is to practice internal skills, some mental skills and internal forces, while in the industrial sector, we practice external skills and pay attention to kung fu tricks. In college, it's a bit like going up the mountain to Shaolin Temple to practice martial arts, and academic masters are like Zhang Sanfeng who founded Wudang School, while starting a company is like going down the mountain, opening a security bureau, and being a product manager is like being a dart, darts, and escorting products to the ground.

Internal and external work complement each other. You do not have the foundation of internal force, and your external skills lack strength, but if you only practice internal skills without external skills, no matter how good the internal skills are, they will rot in your stomach.

When the internal skill is practiced to a certain extent, the real qi moves around the body, running around, and you want to send it out. Maybe you don't do it yourself, your students can do it, too. My personal state is to run at both ends of the mountain. To be exact, there are three places to run: universities, companies, and non-profit organizations (just like when we founded Lianhuashan Research Institute in Hubei).

(the picture shows Zhu Songchun and Shen Xiangyang instructing students at Hubei Lianhuashan Research Institute in 2006.)

Host: it turns out that Mr. Zhu often "goes up and down the mountain". No wonder he keeps such a good figure. For Harry, you have brought a lot of students in industry, so what kind of advice can your experience in industry give them to make their career better?

Shen Xiangyang: actually, when it comes to practicing martial arts, I really practiced martial arts when I was a child. I squatted at Jiming Temple in Nanjing every morning for three consecutive years.

I am relatively lucky. After graduating from my PhD, I have been doing computer vision and graphics research in Microsoft Research USA and Microsoft Research Asia for more than ten years, and have been working on Internet and artificial intelligence engineering and products in the last ten years. I told my students that people are very lucky to be able to learn all their lives. Because most people don't have a chance to do it forever. Even if you are not very good when you graduate, after ten or twenty years of research, you must be very high-because the people in front of you, such as your classmates, are gone, and you become a master. You see, later, I didn't do computer vision research, so Songchun became an expert (laughter).

"going in and out" between academia and industry is a very good thing, and I totally agree with Songchun. You won't know how magnificent the world is until you go to the mountains to have a look. If you go down the mountain and become a security guard, it will be completely different. The industry sometimes feels that the academic community is so fancy that they don't know when it will be realized, so they have to think about it from a different point of view.

Host: it seems that the discussion between the two teachers has produced some sparks. Do you have anything to refute or want to say about Harry's point of view?

Zhu Songchun: it is true that the two of us are from different camps. They graduated from Carnegie Mellon University and are a big "gang" in industry. Brothers and brothers promote each other and are very influential. And I am from Harvard University, my tutor is a mathematician, chemical expert, there are few division brothers. Therefore, I have to walk independently after graduation. When I graduated, when my mentor was having dinner with me, he said to me, "find your alliance (go find your ally)." I met those division brothers of Harry, although they are scholars who meet soldiers (laughter), they are still more polite to me. Of course, Harry is the ally I found, and he takes special care of me.

Shen Xiangyang: just now Songchun mentioned my alma mater, so I'll go on. When I went to Carnegie Mellon University (CMU), I found that the strong students in the United States were really strong, and we did learn a lot in school.

I think every school has its own style, like kung fu, there are Huashan School and Qingcheng School. Our CMU kung fu is more like the Shaolin School, just as Song Chun said that we practice external skills and tricks. For example, the CTO of many American high-tech companies comes from Carnegie Mellon.

As a matter of fact, you can't do a project alone, so when I was a graduate student at CMU, I learned how to organize a large system and bring a group of smart people together. This is what Songchun called an alliance.

Zhu Songchun: Harry is right. Computer vision and artificial intelligence are very complex problems. To build such a system, there must be a strong engineering team. But we must be careful, the water here is very deep, just like the AI-complete problem mentioned earlier. It is risky to start working on a big theoretical framework before it is clear. My colleague Judea Pearl has a saying, "Blind people ride blind horses and cross minefields." At that time, Liu Bei fought everywhere with several fierce generals of Guan Yu and Zhang Fei, but as a result, he was beaten around with little room for a cone. It was not until he met Kong Ming in Longzhong, Hubei, that Zhuge Liang hung up the map and analyzed the situation in the world and the road map clearly before he got on the right track.

Basic research is to provide a large map for the engineering team, and I have just put forward the big pattern, history and map of artificial intelligence in my speech. This "map" is a comprehensive look at all areas of artificial intelligence, so that we can clearly see and think clearly the road map of integration and unity between various fields.

There is such a "big map" in my lab. Although it is not complete and the resolution is not high enough, it provides a guide for graduate students to settle down and do research.

By the way, there are several outstanding young scholars displayed on the wall of today's ACM Turing Conference, which are trained by our laboratory. In the past four ACM lists of excellent doctoral dissertations (two in each year), the winners of three outstanding doctoral dissertations have been trained by our team (Wang Shuo of Peking University in 2015, Liang Xiaodan of Sun Yat-sen University in 2016 and Wang Wenguan of Beijing Institute of Technology in 2018).

On the relationship between mentors and students: two-way choice and tolerance

Host: Mr. Zhu just talked about the relationship between teachers and students. It just so happens that we also want to talk about the relationship between teachers and students. The two teachers have brought a lot of students in the past few decades and maintained a good teacher-student relationship with many students. Recently, a young professor from a well-known university in China was in the news when he was instructing students to write a paper. What do you two teachers think of this? How do you take care of your students all these years?

Zhu Songchun: I think the reason for this is that there are problems in some links. Now colleges and universities require the number of papers for young teachers' evaluation and graduate graduation, but the values of teachers and students are not consistent. The ancients said that "different ways do not conspire against each other". Now many mentors and students have different values, interests are misplaced, and conflicts will arise over a long period of time.

As a teacher, my biggest feeling is that students are often not taught, students are selected. People who come to graduate school are already in their early 20s, and it is difficult for you to change their values and habits. Then you choose students who are close to your values. Students can have a harmonious relationship only if they agree with their mentors' interests and values. The relationship between tutor and student is a process of two-way choice. It is often said that the undergraduate chooses the school, the master chooses the major and the doctor chooses the tutor. The most important thing for you to do a PhD is to choose the right mentor and don't attach too much importance to school rankings.

In the past, the relationship between mentors and doctoral students was a teacher-apprentice relationship and became mentors and mentors after graduation. As the saying goes, you may have more than one spouse in your life, but you only have one mentor. Unfortunately, in today's society, this relationship is no longer so close and has evolved into a relationship between bosses and employees. I do not allow students to call me the boss in private, so they call me "Lao Zhu" behind my back (laughter).

Host: Songchun said well. In fact, teachers choose students, students are also choosing teachers, this is a two-way choice. Especially for students, it is necessary to find mentors who are in line with their interests in the right research direction. Harry, what's your opinion?

Shen Xiangyang: on this issue, I am not as profound and radical as Songchun thought. I think it is very unfortunate to have this kind of news. However, in the field of academic research, when I gave a speech on how to do knowledge in China many years ago, I always stressed one thing: doing research is not everything in life, it is only a part of life, it is an interest and hobby.

I think, as a teacher, it is very important to have love. Every student is different, and there will be differences in creativity, but most of the students we receive have high IQs. I told my wife that taking a student was like having a child. After giving birth, I couldn't return it. What can we do? If you read your graduate student, then I will say no problem, good training. We can be more patient and tolerant.

Zhu Songchun: Harry put it more easily, mainly because he is a part-time tutor, he is not responsible for student graduation, and there is no pressure on the school to publish how many papers. However, on the other hand, Harry and I are lucky to have many excellent students to study with us. We have also brought some graduate students in Microsoft Research Asia, many of whom are very affectionate, which is the greatest gain and pride of being a teacher. But I have also experienced that it is a bit difficult to take those students who are not particularly excellent to do papers.

IV. Career Choice in the Age of artificial Intelligence: accurate Positioning

Host: both teachers are born in the field of computer vision and have made great achievements in the field of research, but they have gone in different directions on the road of artificial intelligence, and now they seem to be coming back. In the course of your respective career development, what experiences do you have to share with young people? What is the relationship between academia and industry as two interactive areas? Especially Mr. Zhu, what kind of message did you bring when you came out to create DMAI, from academia to industry?

Zhu Songchun: with the advent of the era of artificial intelligence, non-computer professionals are worried that their job opportunities will be affected. In fact, for the computer majors here today, artificial intelligence has a great impact on the traditional direction of computer science, such as knowledge expression, algorithm analysis, operating system, programming language, communication architecture, and computer architecture. You need to re-recognize it. Your commonly used concepts and research topics may need to be adjusted. I just mentioned in the lecture that one of the core concepts in ACM is the P and NP problems. In fact, when we study computer vision, our eyes are full of NP-hard. For example, when more than 90% of the people in a country break the law, it may mean that the law needs to be changed. Therefore, under the premise of AI comprehensive steering probability model and random calculation, the discussion of NP problem is not so relevant (relevant).

Now this generation of young students, if they do not want to be hit by AI, can embrace AI and choose to join this trend. Choosing an AI major now is equivalent to choosing a computer major in the 1980s. AI is not only a course, a research direction, its content is very vast. The current career in AI, whether it is to choose to stay in academia to practice internal skills, to go to the industrial sector, even if just want to follow the wind paper, are very good.

In the long run, your career choice depends on how you position yourself, that is, your position and time in the artificial intelligence ecosystem. For example, if you want to open a restaurant, which cuisine do you need to do? is it street snacks, chain stores, or minority cuisine? You need to consider it according to your own interests, strength and surrounding conditions.

If you stay in academia to study, you have to look forward ten or even twenty years. The essence of learning is to go to no man's land. I call this state "the breeze and the moon", which is Su Shi's state of mind when he boated on the Yangtze River at night and thought about life problems. There are too many question marks to explain in the field of artificial intelligence. I began to learn artificial intelligence in the 1980s, driven by curiosity. Just like qu Yuan wrote the question of Heaven at that time, he did not understand many things and wanted to understand various phenomena and the relationship between them. In scientific research, what we need more is to "understanding". As the philosopher Spinoza Baruch Spinoza said: "the highest activity that human beings can acquire is to learn to understand, because understanding is to achieve the freedom of thought" (The highest activity a human being can attain is learning for understanding, because to understand is to be free).

When it comes to R & D in industry, good companies often give you the freedom to be 1-2 years ahead, but now the pace is getting faster and faster, and the time for you to think freely is getting shorter and shorter. There are often "stampedes" under hot spots, and they can't help it. Of course, some giant companies can drink coffee and have a good life, but there is a price for such a comfortable life, just like a frog in warm water.

If conditions permit, if you run at both ends of the academic and industrial circles, you can see the full spectrum of full spectrum, have a deeper understanding of many problems, and make your life more complete and exciting!

Shen Xiangyang: I think everyone's situation is different, especially everyone's understanding is also different, mentality is the most important. You can't think about comparing with Mr. Zhu every day. Mr. Zhu has won a lot of academic awards and won all the awards, then how can you compare with him? Martial arts can be divided into high and low, and so is learning. A real master needs to have a good understanding.

I encourage students whether they want to be a professor, go to industry, or start a start-up company in the future. But what I have always stressed is also a very important thing. No matter what company you go to, the criteria for choosing a job can't just look at money. Be sure to see if you will be stronger in the next three or five years, and whether your personal market value will increase significantly than before. The experience learned is priceless.

05

Fifth, talk about young people's "how to avoid stepping on the pit": deep ploughing and careful work can be done.

Host: ask the two teachers one last question. Now that artificial intelligence has come to the tuyere, for young people, give them some advice based on your experience, so as to avoid stepping on the road and taking fewer detours.

Shen Xiangyang: I think for many of the young students here today, they are in the early stages of their careers. The most important lesson I want to tell you is that apart from being ambitious, you have to do something in a down-to-earth manner.

Just now Songchun also said that it is easy to muddle through, but if you really make up your mind to do something, you must first love it. I believe you must have this state of mind if you can do something great. I have seen a lot of smart students, but did not do great things, because I did not settle down to dig deep in a certain field. Make one direction deeper, and then expand in more directions.

Zhu Songchun: I think this era is both good and bad for young people. Yes, it is because there are so many opportunities for artificial intelligence that my laboratory doctoral students get several offer after graduation, and their starting salary is higher than my salary at UCLA. I don't know, because for young people, especially smart people, they are faced with too many opportunities, easily dragged down by these immediate opportunities, wander among all kinds of seductive opportunities and are brought away, and after a few rounds, they can't find their way, which is a bit like Brownian motion, which I think is a great pity. One thing Harry and I often discuss is that we find that among the students we have brought, the best and brightest students do not do as well as expected and fail to catch up with those who are slightly less qualified but more persistent.

If young people want to be calm, they must be able to stick to their beliefs and do things. If they only do one thing in their life, if they do it well, they will be able to achieve something. Character determines fate, to be particularly tough, Harry also said, thick-skinned, to withstand the criticism of teachers and peers. In particular, smart students should be able to overcome this problem.

Finally, some people have found that in the last 60 years, the development of science lacks a big framework breakthrough, which is different from the great breakthrough era in the early 1900s. According to my observation, we are faced with completely new problems, and we all have to study large and complex systems, such as artificial intelligence, neuroscience and brain science, biological systems, and sociology. Is it true that the very successful reductionism way of thinking in the West in the past needs to turn around and integrate Eastern philosophy and comprehensive thinking? I think this is a question worthy of our consideration.

Host: the so-called reference, analogy, generally speaking, it is also necessary to adhere to a direction long enough to establish a sufficient depth of knowledge and then to the breadth. Let us thank the two teachers for their wonderful dialogue and sharing, and hope to have the opportunity to witness the two teachers'"on Swords" next time. Thank you!

Thank you: I would like to thank the organizing committee of ACM Turing Conference, especially Chairman Liu Yunhao, for their strong support. I would like to thank Hu Jun and Zhu Chengfang for their text editing work.

Https://www.toutiao.com/a6733086464063586829/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.