Sun Jian Computer scientist, a person who helps computers "see" the world, and the former director of Megvii Research Institute. Died of sudden illness on June 14, 2022, at the age of 45.
Text by Yao Yinmi He Qianming Cheng Manqi
Editor by Yao Yinmi
"The computer field is a bit like sedimentary rock. You contribute a thin layer to make the mountain taller. But in the end, people just stand on the top of the mountain, only You can only see what's inside with an The most consistent picture is that computers can “understand” the world as well as humans.
Existing AI applications are already everywhere, from night photography that can be done on mobile phones to real-time translation of conversations, from semi-autonomous driving of cars to factory inspection of defective products on the screen. Like all technologies, the progress of AI is also hidden behind the marketing of product . Large companies hire some engineers who can follow the most cutting-edge research and realize the functions of AI.
These applications may seem light, but in fact they are based on the ingenuity and research of scientific researchers. They are giant data piles accumulated with the help of super computing power and electricity. In the past two or three decades, scientists have been working hard day after day, but they have just moved away human senses before computers can achieve real intelligence.
Vision is the most important sense for "machine to have intelligence". Sun Jian is a very influential scientist in the field of computer vision . He holds more than 40 patents and has published more than 200 papers in top academic conferences and journals. His Google academic citations are as high as 280,000 times, and his H-index index is as high as 121. According to the statistics of the number of citations of papers, Sun Jian ranks among the top five scholars in the field of computer vision in the world.
ResNet, led by Sun Jian, is one of the most important studies in the field of deep learning. In the minds of many people, "one" can be removed. Li Mucheng, chief scientist of Amazon , commented: Assuming you are using convolutional neural network , there is half a chance that you are using ResNet or its variants. The paper has been cited a staggering 120,000 times.
Users may not always realize that the convenience of a product benefits from more precise AI technology. As Steve Jobs said, technology accumulates like sedimentary rocks. Sun Jian, as well as many forerunners and latecomers, spent their lives building the technological mountains higher and laying some foundations for further social changes.
Be the first to answer questions
Sun Jian is a computer vision scientist who grew up in the local area. He completed his bachelor's degree, master's degree and doctorate at Xi'an Jiaotong University . He originally majored in automation and . In his junior year, he was interested in the course " Digital Image Processing ". During the doctoral stage, Sun Jian was jointly trained by Xi'an Jiaotong University and Microsoft. He joined the recently established Microsoft Research Asia (MSRA) and entered the visual group, studying under the dean and chief scientist Shen Xiangyang.
After graduating from Ph.D., Sun Jian stayed at MSRA and worked at Microsoft for thirteen years. Later, MSRA became famous in the Internet field and was known as the "Whampoa Military Academy of China's Internet" and contributed technical executives and chief scientists to many companies.
The early 2000s were the golden age for foreign companies in China, and MSRA was full of talented people. After completing his first year of doctoral studies in Xi'an, Sun Jian came to Beijing for an internship at the New Microsoft Building. There were already researchers and assistant research engineers working in the institute, and “their programming abilities were very good.” Sun Jian saw that the programming path might not work, and “the gap was too big.” Then there was only one way left: do it seriously. research.
However, devoting one's life to scientific research and practice is not exactly the result of a self-effacing choice as Sun Jian described.Sun Jian had a sincere love for scientific research and “had an innocent heart,” his colleague Wei Yichen recalled. Sun Jian once told us that many PhDs never want to write a thesis after graduation, but he always wanted to write and was always able to write. He manages a team of nearly 200 people at Megvii, but there is still a monitor at his workstation that always has his paper on.
Like many fields of science, the important questions of computer vision can only be answered once. In this vertical scientific research universe, there are several core problems that have always been there, corresponding to some goals that can never be solved. Researchers are curious and want to find out “first.”
During MSRA, many colleagues witnessed Sun Jian being obsessed with scientific research and studying papers arduously. He also read on the way to the bathroom. At 11 or 12 o'clock, everyone got off work, but he was still reading. When inspiration strikes while reading, Sun Jian will rush out of the office with his paper, walk through the open workstations, and pull his colleagues over for a lively discussion.
In 2003, smartphones were still the playthings of a few geeks. “At that time, we didn’t even dare to think that this technology could be used on mobile phones.” At that time, Sun Jian had already begun to study how to use software and algorithms to take better photos. That new field of research was later defined as "computational photography," and the primary object of transformation was digital cameras.
conditions are limited, and they are very happy to see the graphics card for special computers that was not seen in good universities at that time. For the purpose of research, Sun Jian and Jia Jiaya pooled their monthly salaries and spent huge sums of money to buy relatively high-end SLR cameras on the market at the time, and then combined them with the MSRA environment to obtain more experimental data. They discussed issues all day long, and many all-nighters were spent every year.
Sun Jian has the most projects at hand, and his colleagues considerately help him order takeout. At twelve o'clock in the morning, as soon as the meal arrived, Sun Jian ate quickly and left quickly after eating. Sometimes when everyone is working on a project, they suddenly find that Sun Jian is missing. When they go outside to the hall, they find that he has quickly fallen asleep on the sofa. When he was woken up, Sun Jian was able to continue doing experiments and writing papers in high spirits. Jiajiaya was so envious that his sleep was not that good.
Later, Sun Jian’s insight became more and more keen, and his status in the MSRA vision team became higher and higher. Jia Jiaya said that many people have realized that Sun Jian is a good listener. He can always find the essence of the problem in discussions and help them open up their minds.
Many years later, Sun Jian and Jia Jiaya have become among the most important figures in the field of computer vision in China. One of them worked at MSRA for 13 years and became the chief scientist of Megvii Technology ; the other joined Hong Kong Chinese as a professor after graduation and served as the co-head of Tencent Youtu Laboratory and later founded Simou Technology .
In MSRA, Sun Jian and Jia Jiaya sit side by side. They are both students of Shun Xiangyang. One talks about ideas and the other listens. They carry the same Swedish THULE backpack and have collaborated on many articles in the field of computational photography. important papers. Sun Jian is the "elder brother" and everyone calls him "Boss Sun".
Later they participated in some activities together. Jia Jiaya and Sun Jian climbed the mountain together carrying several kilograms of SLR cameras and took many photos along the way. Sun Jian always loves taking pictures in life, not only because he likes photography, but also because he needs data. No matter how much it is, it can always be used in the future, he told Jia Jiaya. After
agreed to be interviewed by us, Jia Jiaya wrote a sincere tribute. When the agreed interview time came, he could not read calmly and had to let the AI read aloud. The robot read the nearly 5,000-word article for nearly 20 minutes in a calm tone. Jiajiaya kept his eyes red and left his seat several times. Finally, he simply turned off the camera. Along with the dark scene, the AI read the end of the eulogy, and then, only a few sobs from Jiajiaya were left.
"If you don't do your best, there is no meaning of existence."
Sometimes, a person's high status in the industry may be because he has stayed in this field long enough.However, Sun Jian's status is not the result of simple hard work. Computer vision has experienced many technological changes, large and small. Because of his unique vision, rapid transformation, and continued diligence, Sun Jian has been able to remain at the forefront of technology and lead the industry.
In the ancient times before deep learning, the field of computer vision was relatively niche. Although there were some applications, I hope it was not as extensive and deep as it is today. "If a product cannot pass a technical threshold, it cannot be turned into a product." Face recognition was also an important research direction for Sun Jian back then, but the technical level at that time was unable to enable today's face payment.
In 2012, University of Toronto Geoffrey Hinton's team proposed the 8-layer convolutional neural network model AlexNet. The recognition error rate on ImageNet was only 16.4%, which was nearly 10 percentage points lower than the previous best traditional visual method. It caused a sensation in the entire visual field. field. However, due to the long-term sluggishness of deep learning, there are still many scholars who do not believe in and are unwilling to accept this “new thing”. At the critical moment, Sun Jian resolutely led the team to completely transform. Everyone knows the subsequent story.
Sun Jian’s career and research focus have undergone many successful transformations, from computational photography, to image retrieval, to face recognition, to natural human-computer interaction interfaces, and all the way to deep learning. Every transformation in technology and direction will be somewhat "a waste of martial arts". On the new track, we will fall back to the same starting line as many other competitors and start over.
Wei Yichen knew very well how much courage and determination it took to make this seemingly easy decision. Wei Yichen himself has a deep understanding of this. His doctoral research topic is three-dimensional reconstruction, which is a relatively niche but mature field. In the early days of joining MSRA, I couldn’t even find a suitable research topic. He had been confused for a long time: without transformation, he could not see the direction; transformation meant that his previous experience and knowledge would be shelved, and he was really unwilling to do so. Sun Jian's words awakened him: "Just study for a PhD again." Many years later, Wei Yichen still remembered the inner touch at that time.
Many friends around me know that Sun Jian has the ability to "pure simplicity to simplicity". He is good at using the simplest words to speak the most direct words to people's hearts. The key to making those principles effective is that Sun Jian himself practices and achieves them.
In 2015, the 152-layer residual neural network model ResNet, led by Sun Jian and jointly completed by He Kaiming, Zhang Xiangyu, Ren Shaoqing, participated in the ImageNet Large-scale Visual Recognition Challenge, and the error rate dropped to 3.57%, surpassing it for the first time. The accuracy of the human eye, winning all subdivision competitions. At the top computer vision academic conference CVPR in 2016, the ResNet paper was awarded the Best Paper Award. This was the second time that Sun Jian led the team to win this award. No scientist had ever won this achievement before.
ResNet has been written into textbooks and is a basic skill that every artificial intelligence practitioner must master. Before the emergence of ResNet, researchers suffered from the inability to successfully train the deeper neural network (more than ten layers is already the limit), and thus were unable to give full play to the super modeling capabilities of the neural network. ResNet thoroughly solves this fundamentally important problem. Not only for computer vision, but also for tasks such as speech recognition and machine translation. Later AlphaGo Zero, an upgraded version of the Go artificial intelligence developed by DeepMind, used ResNet.
Perhaps Sun Jian has always cherished the opportunities he has as a technical researcher, "Because technological changes do not happen often, so I happen to encounter them, and I feel very lucky."
He has very high requirements for himself. He once posted on his computer A piece of paper with the words "Only success, no failure" written on it. He also required the team he led "to find the best algorithm and the best solution for a specific problem in a certain field." Moreover, it cannot be static and must always remain the strongest. "Do not do At its best, there is no point in existing."
Technology must be useful
Scientific research is a field that requires all-out efforts. Almost all his life as an adult has been used to accomplish one thing, that is, to conduct scientific research in the field of computer vision that he loves, and try his best to His research results are applied in practice.
MSRA is a research team led by enterprises. Although it has an academic atmosphere, it emphasizes the application prospects of technology. Sun Jian is a very thorough technical pragmatist and left a legacy that influenced him Technical claims from colleagues and students:
“Value pragmatism. Does "
" work? "
" is both a god and a ghost. ”
Sun Jian has become a core member of the MSRA vision team very early. Every year, at the Technology Festival (TechFest) event at Microsoft headquarters, Sun Jian leads his colleagues to find business leaders of the product department to communicate and find technology implementation direction.
Microsoft's image processing software Digital Image Suite has received great help from Sun Jian's team in processing photos. For a period of time, Sun Jian really wanted to incorporate computer vision technology into Microsoft Office software. "It will be there in no time." Available to tens of millions of users. ” Later he really did it.
In 2016, Microsoft released a new cloud service product CaptionBot at the Build conference that year. It is used in hardware devices such as Seeing AI glasses designed for the blind, and can capture the images captured by the camera. The image was turned into a text description, and then voice technology was used to read it to the blind. Later, this technology evolved into a small function of Office Power Point: when inserting a picture, a text description will be automatically generated, and there is no need to manually add text to create a visual image. People with disabilities can read the screen.
He Xiaodong, an AI scientist who worked at Microsoft headquarters at the time, invited Sun Jian to collaborate on this project. He Xiaodong later returned to China in 2018 to join JD.com. He also agreed with the idea of pragmatism in technology. He said: " AI engineering and productization are important points to ensure that AI will not return to the cold winter. "
A former employee of Megvii Research Institute was deeply impressed by Sun Jian's speech at an internal forum. A researcher asked Sun Jian how to balance work and life. The general idea expressed by Sun Jian is as follows: Why should we distinguish between work and life? They should be integrated together. Work is helpful to life, and life can also inspire work.
If someone else said the same thing, this former employee might feel that the company executives were suggesting that employees should "burn themselves" for the company, "But Because it was Teacher Sun who said this." He read it carefully, thought about the truth, and remembered it to this day.
In 2016, Sun Jian joined the AI vision startup Megvii Technology as the chief scientist and founder Print Qi was once his student, Tang Wenbin also worked as an intern at MSRA.
It was an era when scientists were "going overseas" to join companies. After the advent of technological changes, more and more scientific researchers realized that start-up companies had more The opportunities and funds allow them to have more freedom in subdivided fields and apply their scientific research results to the industrial field more quickly.
Many of Sun Jian’s colleagues and students also entered well-known companies or started their own businesses. He Kaiming joined Facebook (Meta), serves as a research scientist at Facebook AI Research Institute (now Meta AI); Cao Xudong founded the autonomous driving technology company Momenta; After Ren Shaoqing left Momenta, he joined Nio Automobile as the Assistant Vice President of Autonomous Driving in North America; Wei Yi Chen joined Megvii in 2018 and served as the dean of Megvii Shanghai Research Institute.
At Megvii, Sun Jian also requires researchers that when applying for a project, they must first ensure that it will have industrial application value within three years before they can start research. He also asked The important person in charge of the project must "go to the field" to see the real environment of technology application, instead of just sitting in front of the computer in the office and imagining an application with formulas and codes.
Many domestic mobile phone manufacturers and products use the technology of Megvii Research Institute. Clearer night photography and more accurate industrial cameras are all convenient for modern people.
Megvii Company has gained huge benefits from Sun Jian. Since Sun Jian took over, Megvii Research Institute’s important scientific research achievements include the AI productivity platform Brain++, the open source deep learning framework Tianyuan MegEngine, the mobile high-efficiency convolutional neural network ShuffleNet, etc.; it has also published academic papers in many top academic conferences and journals. With more than 120 articles, it is one of the companies that has published the most papers in the field of artificial intelligence in China.
Megvii employees all respect Sun Jian, and Megvii Research Institute also plays a very important role in the company. With Sun Jian, Megvii can easily attract outstanding computer vision talents. Many young people join because of Sun Jian.
took over the mantle and forged ahead
After joining Megvii, "Boss Sun" became "Teacher Sun", and this title spread throughout the company. Compared to the "teacher" called "teacher" in most companies in the world, Sun Jian is closer to a true teacher.
Despite the heavy management workload, Sun Jian has always maintained his enthusiasm for scientific research and often personally participated in the discussion and writing of important papers of the institute. Colleagues or students often consult Sun Jian about scientific research or even life topics. Sun Jian always communicates with them patiently and puts himself in their shoes. "I think it has gone beyond the responsibilities of the workplace. Logically speaking, there is no need for that. We are all in a working relationship, so it is really rare. The academic outlook of many young people has been greatly influenced by him." One of Sun Jian's Junior Brother said.
In everyone's description, Sun Jian's image is consistent. He always has a gentle smile. Even if he does not agree with some conclusions, he rarely stands up directly to express his objections strongly. No one has ever witnessed it. Guo Sun Jian lost his temper. Later, colleagues and students who became familiar with Sun Jian knew that silence meant that Teacher Sun had doubts about the discussion.
Whether it is the MSRA Vision Group or the Megvii Research Institute, the researchers there are deeply nostalgic for the simple environment and hard-working atmosphere. Over the years, Sun Jian has silently shaped and maintained this unique temperament.
In the company corridor, researchers whom Sun Jian met by chance would be asked about the progress of the last discussion. Some could not answer (after all, it was only a weekend), and Sun Jian smiled and left, leaving behind gentle but invisible pressure. Later, some students would deliberately bypass Sun Jian's work station, which was located on a traffic thoroughfare.
It is not easy to change and influence people. Sun Jian has a quiet power to moisturize things. "In the early years, my papers were not well written, and I felt hesitant every time I showed them to Teacher Sun." Wei Yichen recalled. "Teacher Sun will try his best to give his opinions. Sometimes you may really not be able to stand it anymore. Teacher Sun will politely say, why don't you go back and think about it." The amazing thing is that after doing this a few times, it was actually written. This is how human potential is forced out.
Sun Jianqian is gentle and low-key. He was less interested in attending public events and more in events for technical people. After Megvii became more and more famous, Sun Jian still turned down most of the interviews, events and public awards invitations. He always said, go for Yinqi, go for Wenbin. Later, he also encouraged young backbones such as Zhang Xiangyu, Fan Haoqiang, and Zhou Erjin to attend events on behalf of the institute and share academic and practical results.
As for him, he has maintained a simple lifestyle for many years, with three main elements: scientific research, sports and food.
As early as MSRA, Sun Jian loved sports very much. Sun Jian and Wei Yichen play badminton together all year round. At the beginning, Sun Jian couldn't even serve the ball. However, not long after, Wei Yichen discovered that Sun Jian's technology had improved by leaps and bounds. Sun Jian used his rigorous professionalism in scientific research to practice badminton. He watched a lot of teaching videos, hired coaches and insisted on long-term practice, summarizing methods and experiences. Later, they won the championship in MSRA's internal badminton team competition.
In the past few years at Megvii, Sun Jian became obsessed with cycling and running again, and even took his classmates from the institute to exercise together. In Megvii’s internal sports check-in group, Sun Jian is one of the few people who can complete at least three sports check-ins every week. The team building of the institute must be related to sports. Sun Jian will take students to climb mountains, run, and experience archery and curling. There are also researchers who suffer from Teacher Sun's enthusiasm for sports and have to make a silent but ineffective protest on the computer - I don't want to lift the iron!
On the forum, Sun Jian’s most published content is not about technology, but about scientific fitness. He has many fitness and sports-related books at home, and he also tells his colleagues how to warm up scientifically before running, how to protect knees during running, and how to stretch after running. A former colleague remembered that Sun Jian once carefully made a PPT to share within the company. He also joked that he can call him "Sun Jianyan" from now on.
In the early morning of June 14, Sun Jian died suddenly due to a sudden illness. Because the news was so sudden, many people around him still couldn't digest the fact well. Sun Jian's work station still maintains its original appearance. When an employee went to work, he placed a bouquet of flowers on his desk.
June 18th is Zhang Xiangyu’s birthday. In previous years, on this day, Sun Jian would send birthday wishes to his beloved disciple early. Yesterday, Zhang Xiangyu came to the company with a bouquet of flowers and bowed deeply to Sun Jian's position. He posted his first answer since registering on Zhihu . He said: Be independent and strong; take over the mantle and forge ahead.
Human history is very long, but human life is very short. A person comes to the world in a hurry and leaves in a hurry. His personal traces may always be healed by time, but his wisdom will become a part of the world.
* "LatePost" interviewed Sun Jian in June 2020, and part of this article is derived from that interview. I would also like to thank Jia Jiaya, Wei Yichen, He Xiaodong, teacher Zhu Mingjie, teacher, and many people who have interacted with Sun Jian at Megvii for their help with this article.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-