The tide of machine learning is booming. How do the tech giants ride the waves? The first round table forum of CCF-GAIR held a hot discussion around this topic. Director of the Microsoft Research Asia Institute, Yu Yu, presided over the opening ceremony. Mr. Huang Jiangji, co-founder of Xiaomi, Dr. Yan Shuicheng, Dean of 360 Institute of Artificial Intelligence, Dr. Huang Feiyue, founder of Tencent, and Dr. Li Lei, today's headline scientist, participated in the discussion. (content is deleted)
Zheng Yu: Share what progress 360 has made recently in the field of artificial intelligence, and what kind of layout will be in the future.
Yan Shuicheng: Before I joined 360, I have done a lot of things in artificial intelligence. Basic visual, speech, semantics, and big data. In addition to the traditional business, there are two businesses that use AI very much. One is smart hardware, and the other is the live broadcast that Old Zhou particularly likes.
From a R&D perspective, we mainly do two things: On the one hand, how do we train this model more quickly after we have big data? In addition, from the perspective of intelligent hardware, in addition to the traditional artificial intelligence, there is a very important scene in the use of the cloud is the artificial intelligence on the side.
There are two levels of artificial intelligence on the side that can do things. One level is how you can make the model better from an algorithmic point of view, reduce complexity to a lower level, and use good artificial intelligence with very few resources. Side; the other is from the surface, for example, to enhance the computing power of the chip. But in China's specific scenario, users are very concerned about the value of the product. In my personal sense, reducing the computational complexity from the algorithm is very valuable, so in the past we spent time and energy in this regard. Much.
For the future layout, I think there are two main aspects: One is the business level, continue to focus on 360 companies' core genetic security business, and provide services with artificial intelligence. From the research perspective, on the one hand, we will continue to focus on building better models and algorithms to improve the ability of artificial intelligence on the side; on the other hand, we will use big data to enhance the ability of deep learning.
360 Institute of Artificial Intelligence Yan Shuicheng
Zheng Yu: The second question, I would like to ask Dr. Huang Feiyue of Tencent. The audience is not very familiar with this department. Please introduce the department first, and what you have done in the field of artificial intelligence and its characteristics.
Huang Feiyue: Youtu is now a division of Tencent. The team was established in 2012. Initially, it was a team under the social network business group. Today, we have spontaneously grown into a department. We are doing research and development of artificial intelligence and related algorithms. With its business operations, it now has dozens of advanced technologies and hundreds of billions of image computing capabilities...
Although excellent diagram is defined to do related research and development, all of our research and development are based on the actual needs of the business, which will form a closed loop with relatively good business and algorithm landing. We can know the good and bad of our business in time and update our algorithm.
Now the team is more optimistic about the direction and business related to the identification of related fields. For example, like biometrics, we prefer the first to face recognition, voiceprint recognition, and then to a series of identifications such as fingerprint recognition: content recognition, pictures, and speech recognition. Face recognition technology, I think it is relatively mature, I believe that in the next few years, face verification has the opportunity to go from the Internet to the offline, this identification scene is more valuable than the current face recognition. Another example is image recognition. Its concepts and areas are relatively broader. I believe that there will be an explosive growth in image recognition in the coming years.
The latest technologies, such as augmented reality and chatbots, are also relatively optimistic. These technologies have truly improved the human life experience. At the same time, they also have a very good combination of offline and online, and they are closer to applications, making it more practical. The value and results.
We hope to really change the quality of life of Internet users by using some technologies of artificial intelligence. At the same time, we will also export some of the latest R&D results through the development platform and Tencent Cloud to ensure that related technologies can truly bring the quality of life to Internet users.
Tencent Yutu Founder, Dr. Huang Feiyue
Zheng Yu: Li Lei is a graduate of the Jiaotong University. He just started to go to Baidu. Now it is the headline today. From your personal perspective, in the field of artificial intelligence, Baidu and today’s headlines have any advantages and disadvantages, especially the lack of clarity. .
Li Lei: I worked for Baidu's Deep Learning Institute for two years before my headline. The content of the work is basically about deep learning and natural speech understanding. After headline, I work in the headline laboratory. The content of the work is about deep learning and natural In terms of speech comprehension and computer learning, before and after comparison, I now do more on the content.
I think the biggest commonality between the two companies is that they are all artificial intelligence-based companies. Baidu no doubt has invested heavily in artificial intelligence. Speech recognition and driverlessness are examples. Everybody's traditional idea is that today's headline is a news client, but now the headline is a platform for all information and content sharing creation. On one side of the platform is the content creator, and on the other is the content consumer. In order to recommend the best content to the most needed readers, what is needed is machine learning technology. From this point of view, both companies are based on artificial intelligence technology.
The difference lies mainly in the stage of development and volume. Baidu agreed with Google and Microsoft to start research in this area very early: Microsoft began speech recognition in 2009; Google used 1,000 computers and 16,000 CPUs for cat face recognition in 2012; Baidu did a deep learning study. The things that are done are voice recognition and image recognition, and almost all areas of artificial intelligence can be done.
The headline today is different from these companies. Since its establishment in 2012, it has only been four years. Can you think of what the big companies are doing when they are established for four years? Facebook has a research institute; Baidu does not have a laboratory; Google is still doing a search. Today's headlines in 2014 had this awareness in the entire team that they wanted to make a huge investment in cutting-edge technology. In the frontier technology of artificial intelligence, no matter whether it can be used immediately, or to do cutting-edge research for technological reserves, the headline concept today is very, very good.
Headline Scientist Li Lei Today
Zheng Yu: Many of millet are artificial intelligence, but we want to know what millet is not artificial intelligence and it is very bad. Because there are many small company representatives here, what you can do badly may be their chance. You talk about what you want to do, but you can't do it well. It's hard to do it. What's the shortcoming?
Huang Jiangji: This is a very good question. Many people say that now is the spring of artificial intelligence. For many startups, it is a good opportunity to overtake the curve. I think this is a fact. But in fact, this corner, like any previous corner, has not changed the difficulty of starting a business. I think if you want to start a business successfully, you still need to understand what the user's pain points are and how to make your products better meet the user's pain points. It is not because you have only mastered a new technology, you can naturally acquire massive users. I think the headline today is a good case, and constantly using new technologies to solve the pain points of users. This is my understanding of opportunities for startups.
Responding to Zheng Yu’s question just now, I don’t think Xiaomi’s skills in the field of artificial intelligence are relatively bad. I think everyone should ask us which of our existing products we have and whether we have a good user experience. , The overall product satisfaction, which is doing worse, I think these are opportunities. If you close the door and ask me, I think that all the products are not smart enough. What we do is just the tip of the iceberg. There is still a long way to go.
Xiaomi co-founder Huang Jiangji
Academic and industry 
Researcher Zheng Yu of Microsoft Research Asia Institute
Zheng Yu: Many college teachers or people in research institutes want to go to the industrial sector, but there are many tricks in my heart. I can survive after I go. So I would like to ask a provocative question. Do you think 360 do? What are the constraints of artificial intelligence or where you are not satisfied?
Yan Shuicheng: I generally explain the problem like to start from the point of view of martial arts. I think that in academic circles, the research methods of artificial intelligence are similar. I hope you are a decent and honest person, defeated others, and can refine some moves. These moves are When others fight, they can win others. In the industrial world, this move becomes less important. What is important is that you have a kind of martial arts, defeat others in a cooperative manner, and solve a problem. This is a good artificial intelligence. .
The problem is that you have to have artificial intelligence thinking and methods to solve this problem. I joined 360 because he has his advantage. In the current special period, we can say that the network is a little bit ahead of the academic world in the field of artificial intelligence. Artificial intelligence resources have great advantages.
When I joined 360, smart hardware was a very critical business. Now that live broadcast is a very critical business, these things are exactly what I want to do. If there are any drawbacks, the pace of the Internet is much faster than that of the academic world. For example, if there are good results in academic circles for six months or three months, there can be a good academic paper, but this is obviously not. Suitable for industry.
Zheng Yu: The Institute is looking more at the future. Has the boss given specific indicators?
Huang Jiangji: There is no standard like the Institute of Microsoft Asia, and a good research institute in China must have two legs at the same time. When you have any results, you need to quickly support existing businesses and let AI algorithms or ideas work in your business. In addition, you need to have a long-term idea, what kind of artificial intelligence technology products need for a year and a half or a year later; or to incubate such business, but it must be able to go online in a very short time. This is a good way to ensure that the institute will survive within the famous enterprises.
Progress and developmentZheng Yu: AI has made great progress in both speech and visual, but there is no great progress in the text. What are the possible technical directions to solve this problem?
Huang Jiangji: My feeling is that the results of deep learning in dealing with natural language and understanding of natural language are far more than ever before, and almost two-thirds of the papers this year were done with deep learning. So it was One premise is that we do not seem to see that deep learning is as successful in natural language as voice technology and image technology. It is not, but it is only a little late. Now there is great progress.
Yan Shuicheng: I personally do more with images and big data. There isn't a lot of understanding of natural language. How people deal with language is different from what we are using now. From a human point of view, the whole direction of the present is different from the way people are, and there are great changes. On the other hand, from an image point of view, if you want to do this kind of image search in general, it is still very difficult, whether Google or Baidu.
Zheng Yu: In the end, each expert summarized the development of artificial intelligence in one sentence.
Huang Jiangji: I think that machine learning may change the methodology of an engineer's product development. It is completely subversive and completely different from before.
Yan Shuicheng: Artificial intelligence makes it possible for people in academia and industry to have a common language. This is very good.
Huang Feiyue: Everyone should not equate artificial intelligence with machine learning. Machine learning is not the same as deep learning. We should not limit our industry to only one issue, and the problem we have to solve is more complicated than supervised learning. We want computers to learn some simple problems, but we must also let computers learn logical reasoning and learn to How to interact with the environment in a complex environment, but also learn how to use the least energy to master new skills.
Li Lei: In a word, it's nothing more than sneaking things. Artificial intelligence is like the Internet more than a decade ago was integrated into our lives. It can integrate into our daily lives and improve the quality of our lives. Our people are unaware of ubiquitous AI.
Nickel Iron Battery,Nickel Iron Alkaline Battery,Solar Battery,Nickel Iron Battery For Solar
Henan Xintaihang Power Source Co.,Ltd , https://www.taihangbattery.com