The GeekPark Innovation Festival was held on January 13, 2017 at 751D•Park Beijing Fashion Design Plaza. The geek park innovation conference has entered its seventh year, and the GIF2017 will last for three days. This may be the most truly global and international “science and technology conference†that cannot be missed each year.
On the morning of the second day of the conference (January 14th), Sogou CEO Wang Xiaochuan came to the scene and shared with everyone the research and progress of Sogou in artificial intelligence, and released the latest overseas search products.
Today we share artificial intelligence for everyone. In fact, this topic has been talked for more than a year. I don't know if you did it. Professor Li Feifei made a 89-page PPT for everyone and told us how much work Google has done in the research community.
Last year, I was actually talking about a sub-heading at the activity of the Geek Park: Artificial intelligence shines into the reality and tells us how we put so many academic and scientific research results into products that we can actually use.
In a year's time, we learned too many words, including AlphaGo, speech recognition, etc. There are many academic concepts. There are also various vocabulary words that we used to call dual networks, anti-survival networks, and so on.
So I was thinking about what kind of context to use to sort out. From study to the end, every consumer, including humans, can face the impact of artificial intelligence on us.
Here we propose a simple model. This model is not only applicable to artificial intelligence or computers, but how we turn technology into a product, how it becomes a commercial value, or the value shared by all human beings, from value and cutting-edge technology. The three dimensions of the user scale.
We often receive résumés, and now there is a particularly tall word for "full-stack engineers." That is, not only can write the front desk, but also write backstage. My own understanding of oneself is to be a "full-stack CEO." It is not only necessary to understand and work from the research, but also to understand the underlying technology, how to make products, and to turn it into commercial value, or the common value of all humanity.
In this case, if you do not have technical content at the beginning, we know that it is very difficult to become a product, but there are technologies that do not necessarily have products and do not necessarily have the size of users. With technology and the size of users, it doesn't necessarily lead to great business value or contribution to the world.
Here I simply made a list, I use a framework to put the existing technology and products of artificial intelligence to put up.
In March of last year, AlphaGo was a black technology. Today we see that it has already been implemented on the technical level and it is already available. However, it does not have user scale and commercial value. AlphaGo includes image descriptions. The image description just described by Professor Li Feifei cannot actually form a user scale.
Internet practitioners, as curious people, can see a lot of amazing technology, but these technologies are not actually used by you. They have not yet reached the second step. Therefore, today in the description of images and the synthesis of images, I don't know if everyone notices them. Like Zhang Peng and Li Feifei, there is a technique of style transfer. I call it the sub-direction of image synthesis. It makes everyone feel cool, but this product, in fact, can not form a user's scale.
Further down, some technologies are not only black technology, but also begin to become user scales, such as voice recognition and face recognition. But these two technologies are in my mind. It is not like machines can make judgments and make choices. In the end, they can form commercial values ​​or bring actual changes to users. The impact is still not big enough.
Can you see what influences it? We as a curious group, we think, can be in the driverless area, this is everyone thinks this is not only feels very dark technology, but also has a large user size in the future, and has great commercial value, or is it to the world change. But I'm sorry I haven’t got it yet.
From this dimension, we will find that today's artificial intelligence has become an upsurge for all people to pursue. However, it has not found a technology that is already available but has a huge user base and tremendous ability to change the world.
This is what we think should happen in the last two or three years.
1942a77qrifwbw0b07k7.jpeg" border="0" alt="" /> This process is difficult, we can make technology, and let this technology stay ahead, it is a difficult thing. Last year, I shared that some technologies are leading when they returned to China. We started to be a company and turn it into a product, but before it became a product or commercialization, this technology began to lag behind and developed particularly fast. Therefore, keeping technology ahead is a problem. What's even more difficult is to turn a technology into a product and benefit a lot of users. Even more difficult is that this technology not only has its user scale, but also can change the world, has great commercial significance and impact on the world.
What I share with you today is some of the trials we are doing.
Before my PPT title on this page was called "AI is crossing the river", but AI is a very general concept. The middle subset is represented by deep learning. It is better to find the law from the answer. Technology, so now the title is called "deep learning is crossing the river."
Sogou, our idea, the first thing is that we have this technology, and we can use this technology to Sogou's products. Many startup companies are trying to construct new products. Our big logic is that we have already used millions of households to use products and then upgrade them.
Therefore, in fact, deep learning has been armed to the teeth by Sogou, and there are a lot of uses in our main products. For example, the input method will use speech recognition and image recognition. With speech recognition, we have now achieved 200 million identifications on a day's scale. At the same time, the amount of voice data that can be produced in a day can be estimated to be 180,000 hours, which is a huge corpus. We probably used 400 GPU servers to support this scale. In addition, when you open the input method, you see that we have the OCR capability. You can use handwriting, or take a picture of the text, and you can do it. So, in fact, we are doing a gradual thing.
Also in the search, I was in Silicon Valley yesterday, and I was communicating with Google's core. Google also said that with deep learning, the user's input can be decomposed. In the past, statistics were based on the statistics of the people we were working on. There were more and more rules and the results were not necessarily good. Relying on deep learning today, the description of the search engine has improved its accuracy by 30%, which is the same as ours. It has become the third most important factor in the ranking of search engines.
This is in-depth learning. We search for this type of company and can quickly push it to use cases. However, such a technology has not yet reached the level of subversion. I feel that it is in line with the emerging world-leading technologies and large-scale users, but it does not constitute a greater impact on the world.
Therefore, we will also do something new here. It is also a great honor for us today to be in the geek park venue to release us with a new product or feature - Sogou overseas search.
Returning to Sogou's logic, we let users express and obtain information more easily. Through the input method, the efficiency of typing in many Chinese people has been improved, and search has given them a lot of information. And just now, through deep learning, this product has become smarter and fuller.
What else can be done to solve it? With deep learning today's cutting-edge results? In this, I share with you the small stories in the Bible. This is of sufficient significance to explain the story of the Tower of Heaven.
As we all know, humans are eager to construct a sky tower through their own efforts and be able to directly speak with God. God felt that he despised his authority. Therefore, he allowed different people and different nationalities to speak different languages. Since everyone speaks different languages, communication becomes an obstacle. Therefore, people do not have the ability to build the Tongtian Tower.
This is often quoted by me. Because of this information asymmetry and non-equivalence, it is difficult for us to work together to achieve the greatest thing. But this is not only about network connectivity, but it also involves one thing, that is, different ethnic languages ​​are different.
Last November, Google had a very important paper that pushed machine translation to a new height. Therefore, in this case, we would like to imagine that if we can express and obtain, we can enable people of different nationalities, including Chinese and English, to freely communicate and obtain their complete information, and this world will change accordingly.
Therefore, in such a mission, I would think that if we can communicate across languages, the progress of such a technology is as important as human beings inventing electricity. Electricity has made our productivity tremendously liberated. However, if cross-language communication can be achieved, the earth’s civilization will be rewritten, the war will become even less, exchanges between different ethnic groups will bring about a new kind of new civilization. So, we did a lot of work here.
On the one hand, under the guidance of Google technology, we have used such technologies completely, based on a translation model similar to the deep neuron network, which we have mastered very well.
And we are fortunate to have more cooperation with Tsinghua, and there are some technical experiences in this area that will soon be integrated into our model. Therefore, it is not only the network of deep learning, but also the synergy of other technologies.
In addition, I think that we are more motivated and determined than Google to do a good job in translating this matter. Google represents the height of technology, but it has a kind of mentality in my heart because it faces not only things like Chinese and English. As a Chinese, we are very much willing to talk to English-speaking nations and English. At the same time, Google is doing hundreds of languages ​​or how to translate between eight languages. I know two languages ​​and can only add the third language, so it is a show of muscle.
However, for the world’s largest language, Chinese, we have a very large motivation to truly turn it into a reality. Therefore, under the drive of such a driving force, we actually achieve product recognition accuracy. It's higher than Google, because we have a stronger desire to do a good translation from Chinese to English, and there are some comparative evaluation data behind it.
Based on this kind of thing, we have already begun to work on our products and have had some black technology. At the World Internet Conference in Wuzhen last year, I started the Sogou machine simultaneous interpretation. Then at the Hong Kong University of Science and Technology, we even brought the machine to the same level of practicality, the process of Chinese to English.
At the Wuzhen Conference we were called the first show because the Wuzhen World Internet Conference was a world-wide Internet conference, but many of the audience here were Chinese. Even foreigners actually wore headphones from the same interpretation. When released is a conceptual product.
When we released this product at Hong Kong University of Science and Technology, about 30% of the students below were really foreign students, and they did not provide other simultaneous interpretation equipment. They directly listened to the machine and listened to it. Afterwards, I asked them if they understood it. ? They said that they did understand what they did in Chinese. I did not advertise the video, but there is some spread online.
In fact, such technology does not conform to the concept I have just mentioned. Although it is a very good technology, it is not for everyone. We put this technology into the Sogou input method at the end of last year. When you were typing Chinese in it, as long as you clicked a button, you could translate it into English. Then click a key, you Can send it out. Therefore, it is easier for you to chat with foreign friends or write e-mails. This is an added feature of our input method. You can see that by clicking on a spline and clicking the translate button, the result of the translation can be obtained.
Today, what I want to bring back to you is that it has been kept secret and because of Geek Park, we have prepared a new product to be released at this event. We call it Sogou’s overseas search.
When we heard the word, we knew that it must be related to cross-language translation work. What is it? When we searched in the past, Chinese information may only account for 10% of the world's information and not more. A lot of quality content is overseas, but unfortunately, one is that Google is currently not available in the mainland. Second, even if we use Google as a Chinese, we really have to admit that our English level is not high enough. There is not enough information when searching in Chinese. It is very difficult to search in English and get answers in English.
Therefore, the overseas search products that we are about to publish today are for the Chinese people and help friends who are not fluent in English or whose effectiveness is not high enough. We have realized that when you type in Chinese words, you can automatically translate them into English to retrieve the content of the English world, translate it into Chinese and read it.
So, the 3 pages of this product can be seen. On the left is the overseas English content of the search. When you click a key, you can read it in Chinese and provide bilingual translation.
Why do you use it? As we have seen before, there are a lot of things that we are interested in overseas, including professional vocabulary, like cellular immunotherapy, and there are inevitably many false information on the Internet so that search can see how overseas people do it. of.
Including, we would like to know the development of artificial intelligence overseas. We can look at artificial intelligence startups. What we search here is what are some of the most advanced artificial intelligence companies in foreign countries, or we want to know the stars we are chasing, lady gaga, everyone. What did she think? We can even see that like Li Feifei, we see her in the Chinese world, and then how do we recognize her in the Western world? Through Sogou overseas search, we can not only search 10% of the world, but also search for 90% of the world. This is (Sogou overseas search) for our huge promotion.
So, we launched this product today. This is the first time that Sogou made such a start at Gekko Park. It is also the first time in the world that the deep neuron network translation system has become a complete search-related product. Can benefit everyone in China. If you give you a translation product alone, it will be very laborious for everyone to use, but we put it in search engines and we will have the opportunity to search the world's content.
We have our QR code on the right. We are very happy to have this opportunity to share with you today. Welcome everyone to try it out.
Behind this, we still have a lot of work to do. This translation work is very delicate, and some frontier vocabularies do not correspond to Chinese and English. Like I came back from the United States yesterday and searched for "United Airlines' online check-in," but translations that have been translated in the past have become United Airlines online check-in. There are many, many tasks that need to be done. Doing a search engine gives us the opportunity to work hard to gradually iterate. At the same time, we need to integrate such search technology with web search. Now it is a separate overseas search, and we must also integrate with our web search. At the same time, we have invested more resources in the machine so that we can have a comprehensive view. We can not only do translation on the search, but also make it possible to read Chinese content in a complete Chinese language. After translation, we can let everyone see the entire world in Chinese.
Therefore, after we have done such a job, if translation technology is resolved, we are wondering what changes will be made to the future. It is also the theme of the conference today. Is it to replace humans or help people? On the Internet, this discussion has begun. Is it to replace humans? Our translation system makes it necessary to throw away professional translations, and the crowd will have no job. However, it has helped more people. Therefore, our repeated question is: With machine learning and deep learning, what is the relationship between machines and people? This is the conclusion of my two reflections:
1. I mentioned "in vitro evolution of wisdom" at the geek park conference in the previous year. After we had the machine, we must be willing to integrate with the machine. We are integrated with the machine. Do not look at people in isolation. We will become more vulnerable to leave the machine. However, if we willingly regard the machine as part of my life, we will become even stronger.
Like a monkey, ask it if you want to become a human? The monkey may be very upset. Also ask a person, do you want to become a superman? We may be very upset. The cost of this superman is a fit with the machine.
What happens if my job is replaced? In the philosophical books I read this past few days, I read another thought of Nietzsche. He said that this earth began to become a large division of labor, and everyone did a partial thing like a screw. For example, translators, drivers, we seem to have found a job, but in fact we have limited our freedom. We have precisely completed an original mission in the social division of labor.
However, when the machine began to replace this exact division of labor, it was actually liberating our bondage. We could do more creative things. Without this loss, there is no new acquisition.
Therefore, there is a saying in my heart that I am willing to be a leading party and bring the intelligent civilization of the machine to the human world. On the other hand, we actually liberated every person who was divided and refined, forcing us to acquire a new kind of freedom and to seek more creative meaning.
This is the relationship between the AI ​​I want to produce and us.
In the end, it took a minute to set aside a few minutes for our videos in Wuzhen and Hong Kong to be done today in simultaneous interpretation.
The speech I gave to everyone today is here. I hope that overseas search will be the first one to be introduced and that can change our lives. It is also an artificial intelligence product that can be used by a wide range of users.
ECG Trunk Cable,ECG Wire,ECG Patient Cable,ECG Lead Wire
Dong guan Sum Wai Electronic Co,. Ltd. , https://www.sw-cables.com