On July 31, 2020, the ECCV 2020 China Pre-Conference hosted by the China Society of Image and Graphics, organized by the Visual Big Data Committee, and co-organized by Beijing Zhiyuan Artificial Intelligence Research Institute and Meituan was successfully held. This ECCV pre-conference upholds the tradition, organized a round table forum, invited experts from the industry and academia, they talked about their understanding and shared some of them around the theme of " Computer Vision: How Big is the GAP of Academia and Industry " Wonderful view.
The roundtable forum was hosted by Dr. Dong Jing from the Institute of Automation of the Chinese Academy of Sciences, Dr. Hua Gang, Vice President and Chief Scientist of Wormpex AI Research, Dr. Tian Qi, Chief Scientist of Huawei Cloud, Dr. Wang Jingdong, Senior Researcher of Microsoft Research Asia, and Wei Xiaolin, Head of Vision Center of Meituan AI Platform The doctor and Professor Xiong Hongkai of Shanghai Jiaotong University were discussing the topic online.
The left column of the above picture from top to bottom is Dong Jing, Tian Qi, and Wei Xiaolin
From top to bottom on the right are Hua Gang, Wang Jingdong, and Xiong Hongkai
Scholars and big coffees have heated discussions on hot topics such as the cultivation of graduate students in the field of computer vision in the industry and academia, the law of talent attraction, and the advantages and disadvantages of cross-industry collaborative research. They also discussed the current industry and academia The gap between the academic frontier exploration of the field and the application of the product, as well as the actual situation of mutual talent flow, conducted in-depth analysis and exchanges, and finally made positive predictions about the future research hotspots and hot industries of computer vision.
Featured views
-----
We have selected 10 expert opinions in the round table forum, and you can see them first:
Graduate students in academia need a kind of philosophical thinking ability.
For those who do research, writing may be very helpful to you throughout your life.
Students need to have a "T-shaped" knowledge structure, not only to have deep research in a certain field, but also to have a breadth of knowledge.
Those who can really make products can discover what society and users need from the perspective of customers. It is also very sensitive to new research directions.
Industry and academia are often a mix, and the close integration of industry and academia is a win-win situation.
Perhaps the industry is like a "market economy", and the academic world can be said to be more inclined to a "planned economy" to a certain extent.
The GAP between academia and industry is from basic research to implementation, which we call the "use gap". The core is that, generally speaking, the academic world is from 0 to 1, and the industrial world is from 1 to n. So, how to connect "0 to 1" and "1 to n" together, and how to shorten this "use gap".
In academia, after your paper is published, the research is over. But it is different in the industry. The publication of the paper may mean that the work has just begun, and there is still a long way to go. The technology must be improved to make it mature enough to support the implementation of the business.
The development trend prediction of computer vision: the interpretability, security, robustness, and transparency of the combination of deep learning and computer vision; small sample training, self-supervised learning, knowledge graph, model compression; data efficiency, software and hardware integration.
Industry opportunities for computer vision applications may exist in the following areas: security, unmanned driving, and offline retail.
Entrance to watch the full video of this roundtable forum:
https://www.bilibili.com/video/BV17a4y177Mf/
Or click "Read the original" at the end of the article to watch the full video
Text record
-----
The following is a transcript of the Panel discussion session, I hope it can be helpful or inspiring for everyone.
Current status and problems in the cultivation of applied/academic talents in the field of computer vision
Dong Jing: Everyone knows that this ECCV pre-conference is for many graduate students in the field of computer vision, including masters and doctors. When they are facing employment, what are their advantages and disadvantages in industry and academia? In addition, from the perspective of teachers, what abilities do you value more of these graduates? What kind of abilities should they have during their studies? In other words, in which direction should these graduate students develop and train?
Xiong Hongkai: As far as the industry is concerned , I believe several other teachers are more experienced than me. I will talk about some of my experiences in academia. I used to have a certain prejudice against computer vision, because the goal of computer vision at the time was to make computers look and feel like humans.
When I was young, I read books and said that we humans face three worlds, one is the virtual world of consciousness, the other is the real physical world, and the third is the world of truth, that is, the world of mathematics. In the past, our academic work was actually more about facing the world of mathematics, looking for and discovering the truths and laws that exist in this world. Later, computer vision opened this window, and I felt that computer vision actually made a certain connection between the human conscious world and the physical world. From this perspective, I think there is a divergence in academic research, whether we should face more truth and laws, or just make a connection between the real world and the virtual world.
Of course, we can say more objectively that computer vision allows computers to create a new species that has the ability to recognize and observe like humans. However, some rational people will also think that computer vision is more of a way of entertainment that paralyzes people. My original prejudice against computer vision may come from this aspect.
But today, we can see that the changes in modern lifestyles are irreversible. Computer vision technology has been extended to many objective aspects, not just in entertainment or consumption. Therefore, I think academia and industry will gradually merge, because after all, the industry is more focused on applications and market development; while in academia, it is now slowly starting from the original truth exploration to humans The expansion of the field is also finding some common points with the industry.
I think graduate students in the field of computer vision, since they are doing research, are bound to face these aspects of thinking. What is the use of what you do? It's not just reflected in economic benefits. So I think that graduates in academia need to have a kind of philosophical thinking ability, not just to do a kind of technology application, nor just to improve some efficiency. I think academic talents must find some real laws and truths that can impress him, not just a simple fusion of reality and virtuality.
Dong Jing: I think Mr. Xiong now puts forward a higher requirement for our graduate students in the field of computer vision. Not only does they require their hands-on practical ability and programming ability to pass the test, but also to rise to a philosophical perspective in discovering and summarizing the rules height. Thank you, Teacher Xiong, for your speech. What does Teacher Wang think about this?
Wang Jingdong: Although I come from the industrial world, in fact, I am more inclined to the academic world. Just now, Mr. Xiong said that the height is very high, so I will talk about some of my feelings and experiences in leading interns with us for more than ten years.
Dr. Shen Xiangyang, the former dean of Microsoft Research Asia, was also Mentor when I was an intern at the research institute. He once mentioned the "three good students", that is, good mathematics, good programming, and a good attitude . First of all, I want to do academic research, including other jobs in the industry. Attitude is the most important. Without a good attitude, just having a good foundation is not enough. At present, especially since 2012, many graduates and doctoral students feel that mathematics is not important, as long as they can adjust the parameters. But when you go to work in the future, you may find that it is not enough to adjust parameters alone, and there is no advantage. Because everyone knows how to adjust parameters, mathematics is still very important. The second is programming. We are in the direction of computer science. Programming is naturally very important.
In addition to the above three aspects, from my many past experiences, including the experience of communicating with students and cooperating with students to do research, I think there is another very important point, that is, communication. The term Communication is actually very broad, not just talking about oral communication, but also writing , such as writing essays, for those of us doing research .
Communication is the most critical point in addition to the above three aspects, because it is difficult for us to do research alone, and we often need to cooperate with many people. When cooperating, such as with interns or students, such problems often arise. When we discuss some issues with students, are the things we discussed and the things accepted by the students the same? In fact, it is not necessarily the same. Then, how to ensure that you can reach a consensus in these discussions, communication (writing) will become very important.
From my experience, it is often recommended that graduate students write more . Writing is not about writing essays, in fact, you must also write about your usual communication. If you write well, then others will know what you think when you look at your stuff, unlike oral communication that will cause a lot of inadequate meaning.
Writing, to help us, not only includes the idea level we just mentioned, but also includes our own planning. It is not only helpful for research, but also very important for future work. For example, it is also very important to do Proposal for teachers in many schools. In the past, Mr. Hua Gang was on our side, and we also discussed this matter together. We did research in the company and lacked a process of writing Proposal. I think this matter is becoming more and more important now. This is from a broad perspective. Proposal is to help you plan what you plan to do in the next few years and how to understand this field.
You may think, I think about these things every day, so can I still not understand this question? But when you actually write these questions, you will find that the gap between (written and thought) is still very large, and the benefits you have brought after writing them are also very large. So, for these graduate students not only are studying, including those who are about to work, whether they are in academia or industry, I think writing may be very helpful to you throughout your life.
Dong Jing: Teacher Jingdong is very demanding on students' pens. They have to write down their ideas so that they can be recognized and seen by more people.
Tian Qi: As you all know, I myself have been in academia for many years. In fact, I am also very close to the academia. I have been a teacher for 17 years. In the past two years, I have also been engaged in some research and industry work in the company.
To answer this question from my perspective is to look at what kind of talents academia and industry need. If you have some of these qualities, then you have some advantages. If you lack them, you may have to make up for them.
Everyone in academia talks a lot, they talk about source research, basic research, and the main research is technological innovation. So from an industrial perspective, I have done some similar reports before, so I summarized a few points:
The first point, speaking from the industrial world, we may need these students to specialize in a field and have a certain depth of professional knowledge, professional skills and creativity in this field.
The second point, in fact, Teacher Jing Dong also mentioned that it is necessary to have the ability to communicate and cooperate, and at the same time to have the ability to understand in multiple fields , which is what we often call a "T-shaped" knowledge structure, not only in this field. Deep, but also the breadth of knowledge.
The third point, in terms of some of our current industry applications, we must have some talents who can really make products, who can discover what society and users need from the perspective of customers, and can think about what computer vision can do from the application scenarios. , In order to stimulate some newer research.
The fourth point is to hope that students will be sensitive to new research directions, including the development of research directions, new scenarios and applications.
Finally, there is another point. Our industry also needs talents with lofty ideals . After all, AI is still people-oriented. Then we should serve people’s basic cultural needs so that AI can better help this society and serve more. customer of.
This is the understanding of student needs that I share from the perspective of the industry.
Dong Jing: I think Teacher Tian actually expressed a point of view that our students should also teach students in accordance with their aptitude and make the best use of their abilities. We can use whatever abilities we have. In fact, we can find "shining points" in industry and academia. "of.
How to treat the enthusiasm of industry and academia to participate in the international summit to publish scientific research results
Dong Jing: This year ECCV 2020 received 1,400 articles, of which only 104 were Oral (oral list), and many of these articles were jointly published by industry and academia. So, what do you think of the enthusiasm of the industry and academia to jointly participate in the international summit to publish scientific research results? This year's ECCV came from a large number of joint contributions from industry and academia. Does it also reflect and reflect the increasingly close integration of the computer vision field in industry and academia? Is there a gap between industry and academia? From the perspectives of useful technology and interesting research, we discuss how we can better help everyone, especially young people, develop.
Hua Gang: The existence of heat is an objective phenomenon. Existence is reasonable. We also hope that this trend can be maintained closely.
The current heat, from a methodological point of view, although many computer vision researchers are unwilling to admit this fact, in fact, computer vision is dominated by deep learning, and one of the limitations of deep learning is that it requires a lot of data. Academia is difficult to produce in batches in the laboratory. In addition, to some extent, computer vision is closely integrated with applications.
From a data point of view, the combination of industry and academia, the benefit of academia is that students can go to the company for internships, or there are some joint projects that can obtain data (industry). From this perspective, the combination of the two can promote the advancement of this field.
From an industrial perspective, the training of PhDs in the past was for academia, and most PhD graduates would be professors. Today, we have "inflation" in all fields. In fact, there are only so many positions in colleges and universities, which creates a profession, and a group of doctors who can do research are also working in the industry.
Therefore, I am here to give you a point of view. In fact, I have been thinking about this question. For a person working in industry, should his work and his Career be treated separately.
Teacher Jing Dong said just now that he works in industry, but he is in academic circles. I do not agree with this view. You are a person in the industrial world, just to say that as a researcher, you have an attribute of academia, that is, you wear different "hats" at different times. Speaking from this perspective, because there are such a group of researchers in the industry, they are also very good researchers. His research has to balance two aspects. On the one hand, he needs to serve the company’s business and serve the business, often his own The energy will be more focused on short-term results, but he himself can continue some longer-term and more exploratory research by cooperating with universities. This is very valuable.
I think this is a win-win situation from two aspects, so I don’t think there is a big gap between the industry and academia , including useful technologies and interesting research, when it comes to computer vision. to today, in fact, very difficult to distinguish between people who are not academics that university teachers and students doing interesting research, and industry researchers just doing useful technology, it is often a Mix . This is my point of view, and I hope that the popularity will continue, because it will be a win-win situation.
Wei Xiaolin: Is there a gap between us now, and where should we go? I think that in the direction of computer vision, the GAP between industry and academia in general has become smaller and smaller, which is obvious to all. GAP is gradually becoming smaller, which is directly beneficial to the industry. Whether it is the recruitment of R&D personnel or the rapid implementation of technology, more and more companies have benefited a lot.
So why is GAP small? First of all, of course, because of the influence of the development of deep learning. At present, the industry has provided great support for the research and development paradigm of deep learning in application scenarios, large-scale data, and computing power, and these scenarios and resources are relatively difficult to reach in the academic world. For example, many papers, especially those of Google and Facebook abroad, are often combined with hundreds of millions of large-scale data and thousands of GPUs, which is unbearable in the academic world.
Another reason is that more and more high-level researchers and professors of computer vision have joined the industry, and the flow of people has also reduced the GAP of academia and industry. Another phenomenon is that in recent years, the industry has recruited a large number of interns in the field of computer vision to do projects and publish papers. Some departments and companies have more interns than full-time employees. This is a "win-win-win" situation. Interns have also accelerated the close integration of industry and academia .
I think that from the present point of view, a small GAP is very good. We don't want industry and academia to have a significant gap. But from the perspective of the future, I feel that sometimes our GAP can be even larger .
I think we have to return to the essence of this matter. What is the ultimate goal of academia and industry? Let's make an analogy. Maybe the industry is a "market economy", and it is more inclined to use market-driven and user-driven to push back step by step, and invert what kind of research we need to do. Academia can be said to be more inclined to a "planned economy" to a certain extent, where the government allocates resources in the top-level areas to help and plan long-term development.
So from this perspective, I actually hope that the academic community can devote more energy to basic research, and can think and do a little further. In this case, for the future, for example, looking at 5 years, 10 years and 20 years in the future, this kind of impetus can actually continue better. So I feel that a small GAP is a good thing now, and a larger GAP is a good thing in the long run.
The importance of industry and academia on top journal and top conference papers
Dong Jing: I hope you can talk about the current industry and academia's emphasis on the top journal papers. Is it the criterion for talent introduction? And the law of attraction of research talents in industry and academia. What is the reason why research talents can come and go freely in academia and industry?
Hua Gang: First of all, I don't want to call the gap between industry and academia a GAP. I hope everyone understands that there is actually a big cycle between research and development and the application of technology to actual scenarios. At the beginning of a start-up technology, you hope that your research can be separated from the development process in stages, and you hope to protect the person doing this basic research so that he has a clean environment to do research.
If you link research with the publication of papers, it means that if the research is done to a certain level, it must be published. If this research is in free form, it can be published in the top journal of the summit. The first difference between academia and industry that I want to talk about is that in academia, after your paper is published, the research is over. But it is different in the industry. The publication of the paper may mean that the work has just begun, and there is still a long way to go. The technology must be improved to make it mature enough to support the implementation of the business.
I define academia as colleges and universities. In fact, the academic community is a large category. It does not only include teachers and students in colleges and universities. As for scientific research talents come and go freely in academia and industry, I think it is everyone's own. There is not much to comment on professional choices. Many people may have the phenomenon of "unacceptable", just to say which industry you are in, which industry you want to do, I think everyone should do it with heart.
As for the degree to which the industry attaches great importance to academic journals, in fact, what research work you have done for the paper itself may not be so important at the stage of job hunting. It is more of a record, indicating that your thinking is logical. Be able to organize your own thinking.
When we are judging interview candidates, if you have these papers, you can add points, but it may only be at the initial "stepping stone" stage, and you may be more likely to attract the attention of the interviewer. In fact, each of my applicants will talk about it by myself to see how many papers he has written, how many ideas are his own, and whether he has formed a mature logic system. I think this is one of the criteria to see if the talent is excellent.
Teachers have mentioned a lot just now, and they pay more attention to the basic skills of talents. I think these are all aspects we will examine. But I think the most important thing is that the talents needed in the industry must have some real skills to solve problems. Because there are many application scenarios in the industry, you still need to chase after the problem and then solve the problem, and build your own technology ( barriers ) in the process .
Tian Qi: I think the core of industry and academia are talents. We all need excellent scientific research talents with solid theory, strong working ability, good cooperation and innovation. At the same time, there are also some differences. As discussed before, the academic community prefers talents who can sublimate practice to a theory. In terms of the industry, it actually means "no matter black cats or white cats, catching mice is a good cat", they value the ability to solve problems.
I just mentioned the importance of the top journals. In fact, our current research institute still puts more emphasis on the top journals, such as CVPR, ECCV, ICCV, etc., which are about the same as the academic circles. If we talk about talent absorption, then we are more likely to focus on the following points: For example, is the author the first author, is the original creator, and whether he has contributed to open source; in some AI competitions, is it possible to show Strong research ability and engineering development ability; whether to participate in very rich and colorful extracurricular activities can improve the breadth of knowledge.
Another is that academia and industry are actually "retro". Nowadays, there is a saying in Huawei that "breaks through the sky". In fact, it is about your ability to innovate. Another sentence is "down to root." "It means that we must stay close to the industry.
Speaking from this perspective, the GAP between academia and industry is from basic research to implementation, which we call the "use gap." The core is that, generally speaking, the academic world is from 0 to 1, and the industrial world is from 1 to n. How to connect "0 to 1" and "1 to n" together, and how to shorten these "usage gaps", in fact, is to study a precipitation process in the middle of landing.
In the end, academia and industry are in a cycle. Just like Hua Gang said, in fact, academia has sent many talents to industry, but many industrial talents have slowly returned to the academic world. It will bring some new opportunities. In this way, it is called complementary and mutually beneficial.
Wei Xiaolin: Regarding the top issue and top meeting, I think it can be said from two aspects. First, the top issue and top meeting is the threshold to enter the industry. In fact, it is very interesting. Everyone has been saying for many years that we should not pay too much attention to the number of papers. We should pursue the essence of the matter, and technology research and development should do truly meaningful long-term research in a pure environment.
But this matter feels that it is still a "dead loop" that can't be jumped out. Whether in academia or industry, in fact, treating papers has become a new "college entrance examination", which is an "open-book examination". The abilities mentioned by the teachers just now are actually reflected in the open-book examination to a certain extent .
So when we get started, we do look at papers, and look at the quantity and quality of your published papers, including whether it is your own ideas. We all value these things very much. However, on another level, we will also be like an "art examination". There are both test papers and interviews. Interviews also play a very important role in exploring the quality and ability of this student.
The second aspect is whether we pay attention to the issue of top publications and conferences within the company. In fact, we are pushing backwards, that is, we still have to push back from the market what kind of technology we need, and when the technology is on the ground There may be some by-products and some outputs, so we will write papers and submit manuscripts by the way, at least in Meituan, because our KPI does not look at the number of top-issue papers.
In addition, when we are talking about industry and academia, there is a very interesting phenomenon that the "academic world" may only have a few academic circles, for example: domestic academia and foreign academia, while "industry" In fact, there are n industries, and every company is an industry, and each company’s market faces different users, so the meaning behind its "industry" is also different. of.
Xiong Hongkai: I think the traditional academic circles, in the 1950s and 1960s, did not actually pay attention to the top journals. I remember they told me that a senior who graduated from Harvard University did not actually publish any papers, and finally became an assistant professor.
Today we always like to give lectures and top magazines. I think this is actually a "lazy" idea, because we are unwilling to judge a person's academic ability, or we can't judge a person's academic ability quickly. Just like our children’s education today, they hope to judge through a certificate, and are unwilling to judge from his actual ideas and content, or to spend a lot of time training and observing him.
I think that in fact, both the industry and academia are facing such a problem, that is, the "fast food culture", and always hope to deal with some things quickly and easily through some certificates or other proofs. Now we take graduate students and recruit teachers in this way. I believe that the industry may also use this method to recruit employees.
Then now it’s “introducing the old to bring forth the new”. According to the acceptance rate of the top journals of the top conferences, determine what kind of top conferences and top journals it is, and finally make a judgment. In fact, there are many good and bad things, and I find it difficult to make a truly objective judgment based on this. This is the same as the college entrance examination. We have selected some outstanding talents, but at the same time we have lost some outstanding talents, so this is difficult to say.
In fact, in academia, many scholars still hope to use traditional methods. It is best to make a judgment based on the person's thinking, whether they are willing to hire him or give him such a development cycle.
Of course, the writing and expression skills mentioned by Teacher Jingdong may also be a key factor. I actually don’t have much say in the industry in this respect, but if you really want to do some academic exploration, or to do some fine knowledge summary and regular discovery in the academic world, you still need to have more independent Thinking, and willing to do exploration, not too fast to do some short and quick research talents. I think this may be a longer-term training method.
Wang Jingdong: Just now, Teacher Wang Liang (director of CSIG Vision Big Data Special Committee) said that I belong to academia, and Teacher Hua Gang seems not to agree completely. In fact, this is a good thing. Today, everyone is not so clear about the distinction between academia and industry. For example, teacher Tian Qi goes from school to company, and teacher Hua Gang goes from school to company. Ten years ago, we were doing computer vision research. In fact, many graduate students did not work in this direction after graduating from a doctoral degree, or it was difficult to find an ideal position in this direction. Today, this thing seems to not exist. A very good thing. For researchers, because in the industry now, there are actually many areas that need to be studied, which is not the same as before after graduation. This is a very good thing, which is why we can switch freely.
So I still encourage everyone, we (industry) still have to do research in the field of computer vision , especially after deep learning, when we do some projects in the industry, if you just say that it is like before If you divide the definition of industry and academia so clearly, the industry can simply realize it. Today, when you are doing industrial work with such a concept, it may be difficult for you to do very well.
The development trend and research hotspot forecast of computer vision
Dong Jing: On the last topic, we would like to invite teachers to share some of their predictions or opinions on the development direction of the computer vision field.
Wang Jingdong: Let me talk about research hotspots first. Although many researchers in the field of computer vision do not admit this matter, deep learning has become the dominant method in computer vision. Deep learning has many disadvantages in computer vision, such as inexplicability, security, and so on. Therefore, how deep learning can be combined with reasoning in the field of computer vision in the future may become a research hotspot, and it is also a problem that many industries need to solve, such as security, transparency, and interpretability.
Hua Gang: First of all, I think computer vision is most likely to flourish in the three areas of the industry. The first one is security. Everyone has seen it, including urban Internet and urban brain, which can all be classified as security. The application prospects are especially in The Chinese market is also obvious. In the second area, I am still very optimistic about computer vision in the field of autonomous driving . The third is that many people have not noticed yet. I think there will be a breakthrough in offline retail . The breakthrough in this industry may happen within the industry, which means that it must go deep into the industry. This is still relatively difficult at the current stage, mainly because the technology has not reached that level, and a lot of investment is needed to break this bottleneck. I think the above three industries are more promising.
In the next stage of computer vision research hotspots, I think there are two trends. First, you will see more and more that deep learning and some knowledge in the field of computer vision are better combined, and the combination will produce better, more interpretable, Some more robust models.
In addition, if you notice, whether it includes CVPR, ECCV, and the proportion of the topics of the submitted papers, you will find that you are actually paying attention to how to use small data to train the model , which can achieve a good computer vision. I think everyone will continue to invest in this area for a period of time. Of course, research hotspots are always difficult to predict, because it is always a quantitative change to a certain stage before there will be a qualitative change, that is, a certain method may suddenly become hot at a certain point, which is unpredictable.
Tian Qi: Generally speaking, there are more talks about the development trend of AI, including autonomy, resource efficiency, security and credibility. Specifically, I think all problems in the field of computer vision are summarized in the extraction of data, models and knowledge.
From a data perspective, we will focus on data efficiency, because deep learning requires a large amount of data annotations. In the future, how can we make the data utilization efficiency better and the deep quality of the data better. From the model point of view, in addition to performance, we have to consider some specific limitations and power consumption delay, which is the design cost.
One of our development trends is to combine software and hardware. At the same time, because computer vision is a technology used, how will hardware manufacturers and AI algorithm manufacturers collaborate in the future? Speaking from a single-point technical direction, for example, for this kind of data efficient, unsupervised learning, AutoML, edge computing, knowledge distillation, and model migration are all good research hotspots in the future.
Wei Xiaolin: In terms of research fields, many teachers have mentioned many different fields just now. When we are thinking about future development trends, there is a big logic behind it, that is, we pay more attention to new development paradigms and changes in research and development paradigms .
Deep learning is a new paradigm that completely subverts the original software system development process. In this broad category, in fact, some new sub-paradigms may appear soon. For example, although the small sample model has not been widely used, its role is to speed up the iteration speed and reduce the computational cost; like self-supervised learning, it reduces the cost and time of manpower labeling and can use a wider range of data; AutoML It can reduce the labor cost of manual research and development; such as multi-modal learning, the combination of vision and knowledge graphs, etc., may even affect the organizational form of the AI department in the long term. In addition, we are also paying close attention to more "cheaper" AI, such as model compression and end-to-side reasoning that bring cost savings.
From the application level, we are more concerned about the new retail that Mr. Hua just mentioned, including the unmanned delivery that Meituan is doing. These are all landing scenarios that can subvert the industry.
Xiong Hongkai: Nowadays, people often say whether artificial intelligence can be completely derived from data, so this means that the big paradigm starts with "brain-like science". Everyone looks at how people think from the brain-like science, or Simply look at the researcher's own thinking from the researcher's own. I see that "neural computing" is becoming more popular nowadays. This belongs to the study of human thinking. I actually have reservations about this, because it may have a negative effect on the person itself.
In addition, I feel that from the perspective of human abilities, humans are more of individuals who integrate the conscious world and the physical world, but the computer world may eventually "combine the two into one", that is, everyone is not. It is possible to distinguish between what is virtual and what is real. Just like the American movie "Western World", perhaps the last need is to enjoy a kind of "virtual" life. This kind of life means that I can put my life through the world, and finally start to enjoy it. Kind of life.
I think if computer vision is based on the final development trend of business, or from the perspective of capital, it may eventually become like this. Finally, it will be integrated with graphics, games, and all real life. This may be the future. However, there must be "two sides".
Dong Jing: Thank you again for sharing. If we make a brief summary on today’s topic, we mainly discuss the GAP between academia and industry, which is actually two directions of a coin. Just like men and women, we have our own strengths and complement each other. It is necessary to communicate continuously to give full play to the complementary advantages. Only in this way can we create a very harmonious and very beautiful future.