I believe that many scholars have had such a puzzle in the face of difficult scientific research topics: When will this material be industrialized and truly benefit the society? Although the belief in the heart will not be shaken, doubts remain for a long time. From the birth of a new technology in the laboratory to the ultimate use of large-scale use, the difficulties and hardships experienced by this will far exceed the imagination of ordinary people. Academic research and commercialization can be said to be two topics that are almost as difficult. Some people are obsessed with exploring the cutting-edge directions and theories of science and technology, but others also hope that they can help bring these technologies to specific products and truly serve everyone. Many of these people eventually went to the relevant company's laboratory to contribute their own strength, and some of them started their own business and began a process of their own technological industrialization. However, the number of dark pits in this area is probably only understood by those who have experienced it personally. In this issue of the hard-to-get public class, we invited CTO Huang Yinning from a polar perspective to answer this question for us. Huang Xiaoning, graduated from the Machine Perception and Intelligence Laboratory of Peking University, and studied with Prof. Cha Hongbin, Professor of Cheung Kong Scholar Distinguished Professor. He has participated in a number of National Natural Science Foundation projects in the field of computer vision and published papers as a key member and obtained relevant patents. He once worked on data mining in Baidu and jointly released a box office forecasting system with the Ministry of Big Data, which was then acquired by Google’s headquarters driverless. It is now a CTO for polar perspectives, responsible for research, design and development of computer vision-related projects, machine learning and deep learning platform construction, and technical architecture design and project development management. Leading the technical realization of CV projects such as CK, China Telecom, Shanghai Park, China Resources Dajiang, etc. Committed to build China's first computer vision PAAS cloud platform, so that CV practitioners in the practical application of the release of their own forces of the original, and promote computer vision technology in the production environment. If you also want to communicate with nearly ten million elite students, and want to become our chief scientist in the industry, please send an email to What is done in the laboratory is computer vision, based on computer vision behavior analysis. Having done ADAS, based on the vehicle speed control of the driving environment, it also does some high-level behavior testing, such as the recognition and quality judgment of common concern, which is an important reference for the early detection of autism in children. Computer vision, machine learning, pattern recognition, three-dimensional vision, compressed sensing, numerical calculation and optimization are some professional courses. There are many choices of courses outside of the profession, and psychology has been chosen. This piece is still quite interesting. Many psychological processes are expressed in body expressions and interactions. Many of these features can be captured by computer vision. In other words, psychology can provide a holistic model, and I translate it in the computer language. At the time, we also chose a lesson about brain structuring and visual formation that was partial creatures, and there was a deep learning mentality there . The development of many sciences is bionics, so we must walk into ourselves and understand ourselves. He also chose Hanfu culture. For a while, he would wear Hanfu in the garden for a while without a weekend. It is quite interesting. Doing data mining is such a consideration. The methodologies for dealing with signals are all consistent, except that the categories of input are not the same. Previously it was done based on vision, but it does not mean that you can only handle visuals with this methodology. Data mining also uses a lot of machine learning and pattern recognition. In fact, it is not exactly the same. At the time, research was conducted on the identification and modeling of high-level behaviors, but it was difficult to identify and analyze behaviors in a truly general environment. There is a lot of experience in visual analysis and behavior analysis. Detection, tracking, attitude, recognition, and understanding of the scene are each a module in behavior analysis. The cumulative noise of the entire system can not be underestimated. However, because these visual tasks in the middle and lower layers are much more complex than the laboratory environment, the noise will be much larger and it will be difficult to achieve common use and commercial use. When some basic algorithms can be used universally, the behavior will be much simpler, otherwise the cumulative error will be too great. If we join the non-research institutes and invest in the industry, I think this condition is quite simple, that is, enjoy the pleasure of making products that can be used. I remember that when the teacher asked me to send Paper, I said to the teacher after I sent it. I think this is not so meaningful. I feel confused. I feel like I'm doing research for Paper, but I don't see some immediate effects. The teacher thinks that the majority of the academic world’s contribution to the world is that someone has seen your research and promoted even a little bit of its border extension . Finally, it’s fed back to the industry and has a substantial impact on the world. And I personally enjoy the thrill of this direct change. In fact, it's okay. I wanted to go. This is a very interesting question, and I want to sample it for a correlation analysis. For example, Caffe author's Jia Yangqing, Caffe is actually a work of his doctoral period, but this framework is very good, not only in the academic community but also received widespread attention in the industry, Google and Facebook to throw him an olive branch is not surprising. And such giants can provide massive resources. You can continue academic research there, and some people will realize your technology. Therefore, this has been a problem one after another. Basically, the academic community has done a lot of work . There will always be people from the industrial sector who come to smell it . For those who start their own businesses, I think they may be like me and hope to make a little change directly to the world. What are the conditions that need to be met if you are not a good student and want to start a business? Knowing that a big god of architecture is a medical student, he may not be good enough in medicine, but he does not want to be the first programmer to be a good doctor. Studying this matter can not make people fully express themselves. I may not be good at computer science, but I would not like to engage in IT entrepreneurial ventures. The key is to explore my own flash point and find a job that suits me. This business is really not suitable for everyone. First of all, you must have perseverance, be optimistic, have confidence, and then you have to work harder. If you choose purely for the sake of wealth and freedom, you still have to come. You have to have a sense of responsibility for the industry identity and love you have created, and you are responsible for the brothers who work together. With these basic qualities, I feel that I can come out and try out. I also need to remember to bring along a team of brothers who will work hard with you. There will certainly be pressure from all parties, and will continue to receive praise and criticism from the outside world. But remember to be optimistic, persistent, and have a sense of mission. When we do research, we sometimes ignore the completion of the pre-step, and verify the methodological aspects of the innovative part of the model to prove that this step is work. But there is no such assumption in the industry. For example, if you do face recognition, the error from detecting the alignment to the feature extraction and retrieving the matching step needs to be reduced as much as possible. This is the project. People will not see how the middle method works. Paying for exquisiteness, looking at the whole . Again in the imagenet battle, you can see that many academic institutions and companies are heaping machines to spell out this precision. But the real product can not be this idea, such as our company is selling services on the cloud, the lower the cost of services, the more acceptable users, after all, China's environment would have been less acceptable to pay for software, and the greater the amount of calculation, pay The cost to the cloud will be greater, this will directly reflect the cost of services. Therefore, we must minimize the amount of computation while pursuing accuracy. Assuming something I know clearly that I can achieve four 9 accuracy, but the calculations must be turned 10 times, we will ask ourselves whether the three 9 or two 9 are acceptable. Industry will only work harder. There are many things to weigh, accuracy, efficiency, product, and model. It is equal to the multidimensional limit condition to find the optimal solution. There is no clear goal in the academic world to go straight to the topic. But I came to the industry to discover that one of the greatest benefits is that I can get a lot of data from actual scenes. After coming to the company, they had a lot of products and felt that it was a very happy and fulfilling thing to be able to use computer vision to help them save labor costs and increase efficiency. One day a park project counted on our passenger flow and felt that it was very practical. It was a lot more reliable than a certain brand on the previous one. We must know that the manufacturer is doing well in this industry. Previously, they were skeptical of the data analyzed by the algorithm and wanted to use it completely. After several random verifications, they completely trusted our data. I know that I am very happy afterwards. This shows that our products are grounded, practical, not conceptual, or arranging artificial intelligence. Of course, being praised by customers is better than competitors. Young scientists are also unable to talk about it. Pits are also unsatisfactory. The methodology cannot be discussed. These words are too big to talk about a few points of experience. From the academic world, sometimes it is very detailed and time-consuming for some things, and it takes a lot of time. It does not mean that this is not good, that is, it will slow down the growth rate. For start-ups, the change response must be fast and innovative. For the product, it does not have to be absolutely perfect again, but it must have characteristics and have core competitiveness. In this period from scratch, the control of small details is not as good as the control of the big trends, and it is difficult to avoid falling into a partial optimum. The first to have to polish. The second is to build a team, which used to be a single soldier or a small gang. All of them were laborers in the lab. Their odors and cooperation were also pleasant. At the company, some people may have good skills but inconsistent goals. Some may be very diligent but can't give output. These are not suitable teammates. When the adjustments are not good, they must be handled decisively. To be willing to spend money on talent, talent knows their value better than others, creating an efficient team can create a geometrically multiple value for the company, and the strong team can cultivate more powerful players. Also understand each person's core appeals, know how to inspire them, and how to help them grow up. The third is to establish a culture and form a certain sense of ritual. The "History of Humanity" stated that communities below 50 can be maintained by word of mouth. More than 50 people will believe in the same story together. For example, we all believe in the concept of the state. So we formed the country. Citizen groups . To establish a good culture, a lot of things can be run on their own, and there is also a basis for initial trust between people. More than that is the challenge of the management level. The difference in technology has been said before. CTO is not a pure technical post, but it still adds a lot of demand. In terms of technology, we need to grasp the direction and take the lead in our practice. Moreover, we must maintain our technological advancement so that we can have a firmer analysis of what technical route to use when analyzing the demand. What is its essence, so I will continue to read paper every week. At the same time, we must keep track of trends in the industry and the demanding population. As a CTO, although it is an algorithm origin, all technology-related matters need to be arranged and managed more or less, that is, doing things without borders, it is difficult to do only the algorithm as before. With regard to management, there will be a lot of information coming into your eyes every day. You need to clarify the urgency and importance of things. There is the need to know people well, to be familiar with each person's strengths and weaknesses, what kind of guidance help or resources they need, to create effective teams, to unite the team, and one team to work like one. Change, change as soon as possible, the premise is really sure that this is not work, it will soon be a mess. I know a person. In the early days of entrepreneurship, I changed directions. I found the direction for the fourth time. I tried quickly and wrongly. However, they are well-funded. Otherwise, the adjustment of the entire staff is more than three times. They are very good now. They have been listed on the C-round. If they are not able to do so without their strong financial resources, they will not be able to do so. To determine whether this is a pseudo-need before sailing, many companies will die from pseudo-demand. Because one of the characteristics of entrepreneurs is optimism. Believe in yourself. You do not believe in how others believe you. But many are blind and optimistic. They think that if they can work, they will start work, but it is actually a pseudo demand. More than 80% of people’s access to information is through vision. The amount of information in images is very large and complex. Before deep learning, academics and industry always knew the value of vision. However, many things only remained in the laboratory and it was difficult to enter the market to achieve commercial accuracy under realistic scenarios. Why do you need to do a deep study when you look at Paper now, not only because of his fire, but because he really turns some algorithms into reality. As long as the visual technology can land, the industry will erupt here, and the visual field has not been sprayed in the industry. It is no surprise that the fire is now. When I was thinking of graduating back then, few companies would recruit computer vision engineers. So many of my senior brothers and sisters went to do natural language processing or data mining. That was the piece of fire. Actually, it was still a little regrettable that I didn't go, but I gained other things, my boyfriend, and now my husband. Everyone is practicing in the world. I once joked with him that if we do not have any external pressure to open a small inn with a wifi near the Bohai Sea, we can read and code in the attic every day and believe that we can make good or even better than it is now. Research. Why do you say so, I think that when you do a thing purely out of interest, it is hard to imagine what the internal driving force can bring out, so the environment is very important and the innermost is more important. As far as the polar perspective is concerned, the CEO has always been in contact with me. I found this paas platform very interesting. When there is hardware in general, there will be a market for software, but it is strange that cameras on the market transmit only video data almost without additional analysis capabilities. 70% of the world's hard drives are loaded with video data. No amount of information is produced in such a huge amount of data, because the lack of brain analysis of these data. The Paas platform is meant to be a computer vision App Store. As long as the camera can connect to a polar viewing platform, I can install an old man's fall monitoring algorithm if there is an elderly person at home. The camera in the shop can install the passenger flow algorithm. It is very interesting, computer vision can solve the problem, more in-depth scenes, these videos are not lying on the hard disk of dead data, can really make sense.
Three Phase Online UPS,Tower Online UPS,Rack Mount Online UPS,Isolation Transformer Shenzhen Unitronic Power System Co., Ltd , https://www.unitronicpower.com
