In the field of artificial intelligence, the term "deep learning" means that the software can improve the performance of the algorithm model through practical experience. For example, in a project launched by Google in 2012, after a neural network running on 16,000 processors browsed 10 million videos on Youtube, the algorithm itself learned to recognize cat faces.
The leader of this project: Wu Enda, has now joined Baidu's Baidu Research Institute and became its chief scientist. Baidu and Google, Microsoft and other search service providers have a natural interest in machine learning. So Baidu established Baidu Research Institute in Sunnyvale, California in 2014, responsible for research in the AI ​​field. It consists of three parts: Big Data Lab, Deep Learning Lab, and Silicon Valley AI Lab.
Promoting collaboration between humans and machine intelligence is a major line of Wu Enda's research. In 2008, he served as a professor of computer science at Stanford University and opened an online machine learning course online. This course eventually evolved into the flagship course of the school's award-winning MOOC online education platform. The number of students in this course has rapidly grown to more than 100,000 in a few years. In 2012, Wu Enda and Stanford University colleague Daphne Koller founded Coursea, a technology education company that provides students with free online courses in leading university education. While becoming the chairman of the board, Wu Enda also retained his position at Stanford.
Wu Enda is one of the leading figures in technology development, and often publicly describes the convenience that smart devices will bring to us in daily life. After the speech at the San Francisco Data Conference in March, Wu Enda was interviewed by Business Strategy (Strategy+business). In the interview, he once again emphasized his views expressed at the conference. In this end-to-end strategy optimized for higher return value, the company's benefits will be greatest, and ultimately benefit the world, both for individuals and for businesses.
S+B: What is deep learning and what value can it bring to a company?
Wu Enda: Deep learning is a new way of presenting ideas that have been developed for decades. In the beginning, people began experimenting with intelligent neural networks that gained some inspiration from the structure of the brain. But until recently our computer has developed such computing power, has such a huge amount of data, and has the ability to help us build some hierarchical concepts that can be learned independently without the design and guidance of human programmers. After the expertise of neural networks. (These neural networks) are beginning to help extract, organize, and process these complex data in modern companies.
A few years ago, we found that deep learning finally began to perform better than the traditional AI implementation. The previous AIs didn't know how to handle the data of this size we are dealing with now.
In the past five years, we have had the computing power to build a neural network that can handle such large amounts of data we have collected from the Internet and mobile devices. By establishing such a large-scale neural network, we can test the performance of our deep learning algorithms. And as we input more and more data into the algorithm, they will perform better and better. This means we can make models and forecasts that are far more accurate than we do now, and use the data we collect to discover problems in the company.
For a small number of companies, deep learning can bring amazing revenue growth. Search engines and online advertising are probably the most important applications for AI in the short term. Search results can be closer to the user's needs, whether it is good for advertisers, users or for us.
S+B: How do you think deep learning will evolve in the next decade?
Wu Enda: I found that the frontier of deep learning is getting closer and closer to high-performance computing. My team and I have built a very powerful supercomputer to take advantage of the vast amount of data we have.
Most of the business value of deep learning in the short term comes from what we call supervised learning. For example, if we look at an algorithm for an email, the algorithm can determine if it is spam. Or we give an ad that predicts when the user will click on the ad. We have created tremendous business value from supervised learning. And we believe that in the next few years we will be able to create even greater and more amazing added value in this.
But looking in the longer direction, I feel that many projects, such as deep recognition such as image recognition and speech recognition, have made great progress on them. Not only will they create huge economic value for the company, but it will also make our world a better place.
Just looking at the development of self-driving cars, deep learning can bring us a safer and more convenient travel experience than it is now, whether it is to increase people's life expectancy or save people a lot of time. In fact, it has given us a lot of life in a few years. According to some data, car accidents may reduce the average life expectancy of children for three years. Americans spend an average of almost three years in their cars. So, to a certain extent, we can say that we have made each person's life more than six years meaningful. This sounds really good.
S+B: You mentioned a self-driving car. Is this really so fast?
Wu Enda: Yes, we hope to make a self-driving car that can be commercialized within three years and achieve mass production within five years. The time period of the final process is hard to predict, but this time is the time we feel most likely to be realized.
For the current state of self-driving cars, it seems that there is an analogy with manual driving that is appropriate. When your car enters a site, it must behave differently than when driving on an ordinary street. It must be slower and pay attention to construction workers.
I don't think that computer vision can reliably distinguish the meaning of construction workers' gestures (such as stopping, walking, and slowing down). But we can solve this problem by appropriately changing the design of the infrastructure. For example, give construction workers a wireless beacon. At the same time, properly adjust the design and construction of the road and social expectations. We can make a driverless car a reality and make it much safer than a human-powered car.
I want to use AI to create a better society by embedding those smart devices in the environment. Today's voice control is like the touch-screen device of the early 2000s. It’s so junior that it’s hard to have any practical value. But with the invention of the iPhone, Jobs and Apple discovered ways to make the touch screen efficient. Voice control will also undergo similar changes. They will affect every aspect of society. Imagine robots and robot guards who follow our voice commands. I hope that my children and grandchildren will wonder why in the future, when we want to adjust the temperature of the air conditioner, we need to turn the dial instead of telling the house directly, "We feel a bit cold."
S+B: You used to work in four large organizations: Stanford, Coursera, Google and now Baidu. As a person who has always wanted to promote the development of deep learning, how do you think we should work together to create a better algorithm?
Wu Enda: I spent a lot of time trying to establish an organization that ensures that all the results of our research can really help enough people. Therefore, we refer to the architecture of an end-to-end research organization such as Baidu Research Institute. For example, if someone invents a deep neural network, then we are responsible for figuring out where the data he needs comes from and how he applies the results to his products, and how this thing ultimately improves the quality of our lives.
At Coursera, when I was most proud of the time when the top faced some tough decisions, there were people other than me who stood up and said, “Let us return to the essence to think and figure out what is the most for learners. Good choice then do that first."
I really appreciate this way of thinking back to the source. There are too many people who fall into one thing and can't extricate themselves, just because he has been doing that or someone else has been doing that. The results of their work may seem more effective, but they may not be real achievements.
S+B: If you want AI and technology to serve people better, people need a system of trust. But people’s trust in the huge system is weakening. How do you solve this problem?
Wu Enda: I think we need to work more closely with the government and society to solve some of the problems related to AI. For example, a very big possible problem is layoffs. As a technician we should be honest with this issue. There are 3.5 million truck drivers in the United States. When the autopilot system matures, what should they do? Similarly, how does AI affect those experts in the medical imaging department? Some people may need to start planning their future.
In general, new technologies will create new, more meaningful roles for people, and this time maybe. However, there may be some problems in this process. This is why I support a minimum (welfare) income for those who can't find a job. But we should ask those who receive these benefits to stay motivated. If we can give those unemployed the skills to get paid, it is a good thing for them and society. The world is changing at an unprecedented rate. In order to keep up with its pace, we must keep learning.
Indoor LED Display is widely used in stadium, shopping mall, station, office and hotel for advertising with characteristics of high brightness, large viewing angle, high safety, long life span, light weight, fast installation and easy maintenance etc. Indoor LED display can be mainly divided into mounted screen and pillared screen. General Equipment for Indoor LED Display: 1. Video processor with multiple signal channel, satisfying the switch of different perspectives from live broadcast, advertising image and on-scene interview; 2. Splicer a kind of high and new equipment dividing a complete image into several blocks to achieve a large dynamic screen with multiple ordinary video units; 3. DVI distributor used for multiple digital video output with one-way digital video input and multi-channel video output function, to display images of the same DVI signal source with multiple digital display terminal in multimedia application system.
Indoor LED Display
Indoor Led Display,Led Advertising Panel,Led Screen Wall,Led Curtain Wall Display
Shenzhen Joy LED Display Co., Ltd. , https://www.joe-led.com