Beijing time on November 14th news, now we can use the neural network in the palm of your hand. Last week, Facebook launched a new tool called "Style Transfer," which adds visual effects to mobile video in real time. After processing, the style of the video is similar to that of the "Simpsons" and Van Gogh paintings. The new tool is to use AI (artificial intelligence) technology to do this. In general, AI requires Dali server to run. Last year, Google introduced neural network technology for the "Google Translate" program. Now Facebook has developed a deep learning system called "Caffe2Go", which is highly condensed and can run on iOS and Android mobile apps.
Recently, Facebook CTO Mike Schroepfer accepted an interview with New Science Man. He talked about the development of the company's AI technology, the impact of AI on communication, and the Facebook News Feed (information flow) received AI. The impact of VR is also affected by AI.
The following is the full text of the dialogue:
Q: To make a neural network run on a mobile device, it needs to be efficient enough. How can it be done?
A: The neural network is the result of a series of steps. The information is processed at each step to provide the content for the next step. From an algorithm perspective, we have a goal: to get the same result, but to reduce the number of steps to The smallest. Developing smaller models and generating similar results is the challenge of the algorithm.
This is the second part of the mobile device optimization algorithm. Even with a smaller neural network model, if you put it directly into a mobile phone, the effect is not good. We must match scientists, some people look for ways to compress models, some people are good at optimizing at the chip level, some people try many different techniques, optimize parts, let the algorithm run faster on the phone, we want to put these people pair.
Q: Modifying the video to make it more like a work of art is really interesting. Can we do other things with AI?
A: It seems to be just for fun, even stupid. Actually it is not. There is an important reason why we pay attention to this: when you develop something new, the delay will turn interesting things into something that is difficult to handle. Because of the time delay, the product is either interesting, creative, or simply unusable.
There are other reasons besides this. We've created some demo samples. In the sample, you can combine the application with the target detection. If you want to add different effects to the foreground and background of the video, you can do it.
Q: Where is the neural network for Facebook training used?
A: Used in many different places. For example, in translation, users upload billions of pictures every day. We use neural networks to add captions to pictures. If you have problems with your vision, I hope the pictures can tell you what it is, and the neural network can be implemented. When sorting information flow, we also use neural networks. You may see thousands of stories, but only 10, 20 or 30 are really read. We want to show you the best stories. . We also use neural networks to detect spam. If someone shares a link on Facebook, the link is no longer valid, we can find it and then delete it.
Q: You have said before that VR will play an important role in social interaction. Can Facebook AI help?
A: To make VR practical, AI is a key technology. In the real world, we need to figure out the position of the head and hands and draw their positions in the VR world, which involves computer vision and VR. Without them, the system will not run. Today we can achieve it, but in 10 and 20 years, you can hardly do it.
We might as well consider a more in-depth question: how to put a real substitute into the VR world. For example, we are in the VR world with someone, he is laughing, I must detect his smile, and the avatar in the VR should also laugh. When someone speaks, we should analyze the language expression, let the VR substitute's mouth look the same as the real person, and talk like a real person, instead of no reaction, no action. If the VR substitute is indifferent throughout the process, we will lack a sense of presence.
In the long run, all future systems will be built on intelligent agents, which are information robots or things that can be exchanged at home. It's a good idea to put these features in a VR environment, because they can navigate you and let you navigate through many virtual worlds. You can say, "Hey, take me to Mars." Or say: "Take me to see my friend Joe." The virtual agent will take you when he hears the command, without having to click the menu, no need to move the button. For virtual assistants, VR is a very suitable place to exist, just to achieve some remote.
Q: How can I achieve this goal?
A: In the AI ​​field, the problem of speech recognition has been well solved. This technology is already very good, but there is a bigger challenge in the AI ​​field: improving natural language understanding and eliminating ambiguity. When I gave the order, "Take me to Mars." What does it mean? Is it a certain game? Is it a trailer in Mars Rescue? What do I mean by that? This is a problem for AI.
If the system is very practical and knows what I want, it looks quite powerful, like magic. If the answer given is wrong, it can be frustrating. Because of this, we have to develop a system to make it more practical, otherwise people will not want it. For AI, this is a problem: develop a system that allows it to understand the language in a human way.
Q: Wait a day if we can put the neural network into the pocket, what will happen then? Can you describe it?
A: There is a resource that is precious. Once it is lost, it can't be obtained. It is time. Day after day, year after year, time lapses and can no longer be returned. With AI, we can spend time on things that we care more about. I can spend more time learning 3 languages ​​to communicate better with my family. If there is a system that can be translated automatically, I don't have to learn. I will have more time to get along with my family. I can also use time. It's fine to create music, pursue personal hobbies, or work and do whatever you want.
I have a hope that people will no longer waste time on things that are not important, because we have a system that can look after us and let us focus on what we care about the most. (Compile / Tiger Tao)
RAM/RFM Middle Frequency Water Cooled Capacitors
RAM/RFM Middle frequency water cooled capacitors
Water Cooled Capacitors,Medium Frequency Capacitor,Intermediate Frequency Capacitors
YANGZHOU POSITIONING TECH CO., LTD , https://www.yzpstcc.com