On November 17, Shi Feng, vice president of Facebook, said at the 2016 World Internet Conference that artificial intelligence has been integrated into all aspects of their products. They hope that artificial intelligence can help all people create better videos on Facebook. It is not limited to professionals. One of the most important technologies is video stabilization technology, so that even ordinary people can be more professional.
Shi Feng said that if someone tries to fool the camera, no matter how fast he moves or changes his body shape, he can quickly identify and track the person's activities. Just like when the photo recognition was first introduced, (system) has reached 97% accuracy of face recognition. Now it is better to recognize faces than people, and it also starts from face recognition in video.
He also mentioned that Facebook just started testing new technologies in Ireland two weeks ago, and that mobile phones will be able to run artificial intelligence, which is to launch a neural network in the Facebook application, and the computing is done by mobile phone.
He said that these are mostly existing AI applications that cannot be done in the fields of image and video recognition and automatic translation. “For artificial intelligence, it takes us a long time to train it, just like teaching children to play baseball. On the one hand, we have more than one billion people looking for content, and on the other hand, we have hundreds of millions of content published to them. They can understand how the two balance, so we make a lot of predictions every day, and the real excitement is that we can get a lot of feedback," Shi Feng said.
The next is the Shifeng speech record:
good morning everyone! I am very happy to participate in the World Internet Conference for the third time. I come every year, but my Chinese is still very bad. I am not embarrassed. I still speak in English. In our work, we feel that the most important thing is artificial intelligence. Next, I will introduce you to the development of Facebook in artificial intelligence. It can be said that artificial intelligence has been integrated into all aspects of our products. Today we give you many examples. For example, artificial intelligence supports our hardware and software. The last part introduces our recent research results. The content is how to teach machines. Go read the text.
For example, let's talk about news push. Let me briefly introduce you to the fact that 1.2 billion users visit Facebook website every day. It can be said that they spend a lot of time here. For our company, our job is to serve this ten. 200 million people, we have provided them with many stories, and we have provided them with a lot of content. For us, our job is to understand what the user really wants and to push it to them.
For example, we understand each user's interests and provide them with relevant content. For example, if I am in Wuzhen, if my wife and children send photos, the content is that they are climbing Huangshan. I definitely hope that they are the first one of my news push, because they are the people I care the most, and their mountaineering activities. It is also my interest. I am particularly happy to come back to Wuzhen today. I still remember that Wuzhen was also very exciting last year.
From the perspective of artificial intelligence, I think everyone here knows that artificial intelligence is different from traditional computers. It is more like a child learning the world than programming in advance, so it takes a long time for artificial intelligence. Training it is like teaching a child to play baseball. On the one hand, we have more than one billion people looking for content, and on the other hand, we have hundreds of millions of content published to them. They can understand how to balance the two, so we make a lot of predictions every day, and really excitement. The heart of the people is that we can get a lot of feedback.
Once we post content to users, we know what they like, what they share with their friends, what they click, what videos they watch, and more. So we get a lot of feedback every day, and in the process we can also help us train the system. Therefore, if everyone is doing artificial intelligence, everyone must feel that this is an ideal situation. In addition to this, there are two things that help us succeed now. We returned to 1992, when I just completed my Ph.D., part of my research is artificial intelligence. Personally, I think 1992. It's still close at hand, but in fact we look at computer resources in the process of leap.
It can be said that if we leave this leap, we certainly have no way to deal with such a large amount of data. And it can be said that what is even more exciting is that we have made real progress in this regard. Everyone looks at this PPT. You can see the results of the smart competition on this chart. On the left side, you can see the accuracy of the system. The content is the identification image. You can see that in four years, the accuracy rate is from 55. % has risen to 80%, so it is a great improvement. When you look at these circles, you can see the computing power. What is particularly exciting is the recent system. You can see that the computing power is greatly improved compared to four years ago. In the next part, we will talk about these technologies for our daily activities. What is the impact? You can see here are some examples of how to parse images on Facebook and incorporate artificial intelligence into all of the content.
Next, I will introduce some pictures and videos. I think this is part of the most cutting-edge technology. I think there are a lot of Facebook users who have some obstacles in their vision. If they want them to recognize the pictures, they used to do it. No, now we can help them, for example, watching the video here, we can also let people analyze these pictures. If you are blind, this is undoubtedly a big development and progress. In addition, we know that a piece of content on Facebook is growing very fast. It is a video. We didn't release the sound here, but here we can see the automatic translation we performed. You can see the text at the bottom, which means the sound material. So even if you don't open the sound, you can still see the subtitles and know what the following is.
We hope to help all people create better videos on Facebook, not just professionals. So one of the most important technologies is video stabilization technology. We use artificial intelligence technology so that even ordinary people can More professional, and it is also developing rapidly in this field. We have always been people-oriented, driving technology, so we say that our technology is most concerned about people, here is a developer, he is trying to fool the camera here. The speed of their activities, the speed of this R & D staff is very fast, and constantly changing their body posture, and wearing clothes, we know that the movement of clothes is different from people, and sometimes adjust the color and background color. But even so, the system can quickly identify and track the activity of this person. It’s like when we first introduced photo recognition. In two and a half years, we have achieved 97% accuracy in face recognition. Now we are better at recognizing faces than people, and in video we start with face recognition.
Here are some other examples, and our breakthrough, this is the new technology we just started testing in Ireland two weeks ago. You can see that for the first time in history, you can run artificial intelligence on your mobile phone. What exactly is this technology? We have launched a neural network in Facebook applications. You just take a photo with your mobile phone, then you can choose one. The famous art genre filter, followed by photos, can form the final result, and these are real-time. Why are these very interesting, because all the processes and operations are carried out in the mobile phone. In general, in the past, you need to access the cloud to gain computing power to complete. We are still training our system in the original way, but we are now faster and only need the hardware of the phone to complete the original operation.
In order to further develop technology and advance artificial intelligence, we are constantly inventing new hardware. This will help us process more data faster. I am also very happy. We are all open source designs. Recently we also Providing support to nine different countries and 19 financial institutions, so that these industries are making faster progress, we provide them with our specially designed hardware, and we are very happy to help all aspects of progress.
I just gave you a concrete example, telling us about the technology in the product. Next, I will introduce you to the research field that I am most interested in. Our system can tell you very effectively that there is a pizza on the left and a pizza on the right. The person in glasses holds a banana in his hand. But there are two things that computers don't do well. One is to understand the situation and the text, and the other is production. For example, on the left we ask the computer whether this is a vegetarian pizza, the computer is difficult to answer. The right side asks if this person has 2.0 vision, and the computer has no way to answer. People can answer these questions. Because you will soon know this problem, meaning that vegetarian food is not foraging. At first glance, you know that it is definitely not vegetarian. On the right side, this person wears glasses and knows that this person's vision is definitely not 2.0, but the machine can't do it. What can we do to make the machine understand the text? We must start to deal with this problem.
Easy Electronic Technology Co.,Ltd , https://www.pcelectronicgroup.com