Electronic enthusiasts eight o'clock: As the top international conference on artificial intelligence, the AAAI conference organized by American AssociaTIon for ArTIficial Intelligence was held in San Francisco, USA. At this year's conference, the Chinese face became a force to be reckoned with. In 2,571 submissions, China and the United States contributed 31% and 30% respectively. Although China is still lower than the United States in terms of the number of papers received, the number has increased significantly.
An episode is that this year's AAAI conference was originally planned to be held in New Orleans. Due to the conflict with the Chinese New Year, AAAI Fellow, AAAI's current Executive Committee Professor Yang Qiang and several professors urgently sent emails to the organizing committee, which eventually changed the situation. time and place.
It can be seen that China is gaining momentum in the AI ​​field and gradually growing into a backbone. According to the "Wuzhen Index: Global Artificial Intelligence Development Report", in terms of the number of global artificial intelligence patents, China ranked second with 15,745 followed by 26,891 in the United States, and Japan ranked third with 14,604. It is worth mentioning that the three countries accounted for 73.85% of the total patents.
Chinese forces"Not only scholars, but also the number of Chinese companies attending the conference has increased." Yang Qiang said. From this year's event sponsors, Baidu, Tencent and Amazon, IBM have become gold sponsors, small i robots, today's headlines are also among the silver sponsors. In the papers included this year, the artificial intelligence teams of Chinese companies such as Baidu, Tencent, Huawei, 360, Today's headlines, and Ctrip also appeared.
Lin Yuanqing, dean of Baidu Research Institute, told the First Financial Reporter that in the past few years, she has seen a lot of Chinese attending the conference in the top conferences in the field of artificial intelligence in the world, and it has grown very much in the past few years. fast. He believes that this is also related to the investment of several domestic companies in the field of artificial intelligence.
“Chinese people are suitable for artificial intelligence. 43% of the world's artificial intelligence papers are written by Chinese people.†Li Kaifu (microblogging), founder of the innovation workshop, once told the First Financial. According to the National White Intelligence Research and Development Strategy Plan previously published by the White House, from 2013 to 2015, the articles of “deep learning†or “deep neural network†increased by about 6 times in the papers collected by SCI. According to the number of articles, the United States is no longer the world's number one. After adding the "articles must be quoted at least once" additional conditions, China surpassed the United States in 2014 and 2015, ranking first.
"This round of artificial intelligence is not a new revolution, but a continuation of the automation of the industrial revolution in the 18th century. Once the technology is in its hands, it can be rapidly expanded to do business all over the world, so this has played a corner overtaking for China. The role." Yang Qiang said.
Chinese mathematics and hard work have undoubtedly provided a good foundation for China to develop artificial intelligence, but the greater driving force lies in industrial demand. On the one hand, for traditional enterprises, new technologies are needed to promote industrial transformation. "China's economic structure has many unreasonable and inefficient places. Through the wave of artificial intelligence, a new kind of competition has formed." Yang Qiang stressed .
For the Internet giant or the emerging unicorn company, it is also necessary to use artificial intelligence technology to stimulate the massive data already stored, improve service accuracy and create potential profit opportunities. “The application of the Internet big market to the C round requires artificial intelligence. Li Kaifu said. For example, today's headlines use artificial intelligence technology to reorder news content and video, and realize thousands of information distribution. Meitu also uses portrait database to mark and structure data and optimize image algorithms.
"There are only a large amount of data, large-scale computing and application scenarios in China and the United States in the world. At the application level, China and the United States are basically at the same starting line." Yu Kai, founder and CEO of Horizon Robotics, told the First Financial News. Yu Kai has served as Associate Dean of Baidu Research Institute and Director of Deep Learning Lab (IDL). He led the team to successfully apply deep learning technology to advertising, search, image, voice, etc. Before that, he also studied in the US NEC. The Institute, Siemens Data Research, and Microsoft Research Asia.
In Yu Kai's view, China has the world's largest Internet company, and has good application scenarios such as search, social, e-commerce, Internet finance, etc. "A large-scale computing platform requires large-scale application scenarios, in small experiments. The room can't do it. Young people will get continuous training in such a work environment, including engineering experiment ability and understanding of algorithms."
“The biggest advantage is that there are many people. This advantage is reflected in three levels. People mean that the market is big and there is a stronger driving force to do this well. Secondly, for the social service level, it needs a lot of data. Third The talent base is relatively large, and there are relatively more top talents.†The fourth paradigm founder and CEO Dai Wenyuan told reporters: “From the perspective of data volume and human capital invested, there is not much difference between China and the United States, and China have more advantages."
Sino-US differencesHowever, it is unfair to regard the number of papers as the level of artificial intelligence development in China. Although it is close to the US in the application of commercial value, in terms of basic, original research, innovative soil and talent reserve, China is still relatively small compared with the United States. difference.
"More domestic technology is the landing, industrialization and application of technology. There are still many people in foreign countries doing frontier research in companies and research institutes, including seeking methodological breakthroughs. We are good at making things more detailed, relatively breakthrough and The ground-breaking work is not enough.†Huang Chang, co-founder of Horizon Robotics and vice president of algorithms, told CB.
Huang Chang graduated from the Department of Computer Science and Technology of Tsinghua University. He was a researcher at the University of Southern California and the American Academy of NEC. In 2012, he joined the Baidu US R&D Center. In 2013, he and Yu Kai participated in the establishment of the Baidu Deep Learning Institute as a senior scientist. , Director R&D Architect. In Huang Chang's view, doing research is nothing more than finding new problems and researching new methods. In these two aspects, there is still a big gap between domestic and foreign countries.
Yang Qiang believes that deep learning is constantly evolving, and leaders in the research field should develop new fields instead of digging deeper on the original basis. "Expanding a 10-layer depth model to 100 or even 1000 layers, I think this is indeed an improvement. The Chinese are currently at this level, but these are not original in my opinion." Yang Qiang said.
"Now many colleges and universities are looking at the essays of professors and students. The publication of the top conference papers is helpful for students applying for colleges, teaching ratings, applying for research funding, etc., truly making breakthrough theoretical research, not meeting the assessment system very much. Less." Dai Wenyuan bluntly. In his view, although a considerable number of people are involved in artificial intelligence research, excellent research results are not directly proportional to the surge in the number of participants.
Yu Kai believes that some Chinese students are very good at "scoring points" and "swapping the list." "Others do 99.5%, I did 99.7%, and there is not necessarily a substantial breakthrough. The world has not become different because of this brushing. Original innovation requires different thinking, and now more depth learning. All people are doing deep learning, not thinking about What is wrong ? How to be different?" Yu Kai stressed.
Dai Wenyuan, who has been in the field of artificial intelligence for more than ten years, has the same feeling. "Many people have problems with the direction of force. The accuracy rate is 99.1%, 99.15% or 99.2%. In fact, there is no difference. It should not be used in these places. Instead, focus on the field of less than 60 points and pass it."
Looking back to the historical development of deep learning, it is the path of a marginalized project to mainstream technology. As early as the early 1980s, Hinton, a pioneer of the deep learning school, insisted on the exploration of neural networks, but limited by the speed of the computer at that time, the amount of data, etc., deep learning theory is a marginal study, at that time AI The mainstream research direction is completely opposite to it, and it advocates small sample learning and mainly promotes SVM learning.
It is the insistence of a group of people represented by Hinton on deep learning that has gradually turned the edge issue into the core technology of artificial intelligence. "After entering the field ten years ago, Chinese students are learning optimization theory. Now they are learning deep learning. Few people are wondering whether deep learning is the optimal solution. Just like few people have thought about whether optimization is the optimal solution. Said Dai Wenyuan.
High staff costsIn Yu Kai's view, the gap between China and the United States is manifested in two aspects. On the one hand, there is a shortage of talent reserves. Many colleges and universities do not have artificial intelligence professional for a long time, but there are artificially large institutions in the United States. Intelligent professor. Take Cameroon University as an example, there is a specialized robot research institute, of which there are more than 100 professors of light. In vertical terms, the time for Chinese layout is relatively late.
As early as 2012, Yu Kai returned to China to set up an artificial intelligence team in Baidu, and served as the executive dean of Baidu Artificial Intelligence Research Institute. In his memory, it was very difficult to recruit people in colleges and universities. Many of them were trained after recruiting Baidu.
Secondly, from the perspective of the industry chain, Google (microblogging) or Facebook's artificial intelligence team can not only recruit people from Stanford and other institutions, but also can dig away talents in the field of artificial intelligence from large companies such as Microsoft, IBM, HP, etc. The company also wants to dig people from Baidu, whether it is from research education or the whole industry, it is late and the scale is still small."
So far, Yu Kai still frequently goes to the United States to participate in some academic conferences, so that he can keep more thinking. "There are many foreign technology entrepreneurs. Everyone is discussing mathematical formulas and algorithms. In China, most of them are talking about trends, concepts, and if PPT. Putting on the formula becomes very boring, and the mentality is rather impetuous."
Driven by capital, artificial intelligence has become the hottest area of ​​entrepreneurship and is accelerating the flow of talent. According to the “2016 Early Corporate Salary Survey Report†released by Huachuang Capital, early-stage companies in the field of artificial intelligence and big data have experienced a turnover rate of 44% in the past year, and the turnover of employees has been active.
“Cannot afford wages and grab people†has become the biggest game that artificial intelligence companies face in recruiting talents. “There are fewer talents and more companies, and the cost of artificial intelligence is so high.†Dai Wenyuan said, “We want to find talents who break through the routine and need to find something that can score 30 points or even 80 points. There are many people, such as those who are currently doing deep learning, but there are very few people who migrate and learn."
"The combination of virtue and political integrity" is the standard for Yu Kai's selection. The so-called morality is the enthusiasm for artificial intelligence itself, and is willing to do long-term struggle for it, not short-term. "Most people are fashionable, if the heart is still hot when it is snowy, it is called enthusiasm", it is the mathematics foundation, statistical foundation, programming ability and so on.
"Excellent talents and high-quality research results are always scarce. For example, the number of papers in the field of artificial intelligence has risen from 800 articles per year to 3,000 articles. However, the number of truly outstanding papers will not change much in many cases. Many people are following the trend. Digging holes and watering, solving the problem of fine details, the real value is not great." Huang Chang added.
Different from O2O, e-commerce and other industries, the technological innovation of artificial intelligence still needs long-term and basic theoretical research work. How to proceed from the top-level design, strengthen the basic theoretical research and core technology breakthrough of artificial intelligence, and strengthen the artificial intelligence research talents and technical talents. The cultivation and introduction is the continuous driving force for the development of artificial intelligence.
Artificial intelligence challengeUnder the prosperous situation, it is more important to face the role of artificial intelligence. "Compared to telling people what artificial intelligence can do, it is more important now to tell people that artificial intelligence can't do anything." Yu Kai said with a smile. Combining the current developments, artificial intelligence still faces many challenges.
The first challenge is the problem of insufficient data. As we all know, artificial intelligence is built on the basis of massive data, and the algorithm model is optimized through big data training. Taking face recognition technology as an example, training this algorithm model requires at least one million levels of image data.
At present, artificial intelligence is mainly supervised learning, and supervised training requires tagged data, so the quality and accuracy of the data and the output are closely related. “How to eliminate noise and spam in data and obtain high-quality and tagged data has become a new challenge. It is for this reason that semi-supervised or even unsupervised learning methods will become a hot research topic in the future.†Huang Chang said .
Another major challenge is the lack of deep learning and the ability to migrate scenarios. Data in each area needs to be re-collected, standardized, and retrained, making it difficult to promote across domains. These challenges are also an urgent need for breakthroughs in the artificial intelligence industry and academia. “In the process of recruiting, there are many people who study deep learning, and there are very few people who know how to migrate and learn, and there are few people with speculative ability.†Dai Wenyuan said. Reflecting on talent development and education, it is especially important to guide and encourage students to conduct cross-disciplinary and original exploration research.
For example, this year's AAAI best paper comes from Russell Stewart and Stefano Drmon of the Department of Computer Science at Stanford University. Their paper "Using Physics and Domain-Specific Knowledge to Make Neural Networks Unsupervised Learning" is to physics knowledge and deep learning. Combined, cross-domain research brings new inspiration to AI.
Disclaimer: The electronic reprinted works of E-Commerce Network are as far as possible to indicate the source, and all rights of the owner of the work are not transferred due to the reprint of this site. If the author does not agree to reprint, please inform the site to delete or correct it. Reprinted works may be subject to change in title or content.
Brake Disc For HYUNDAI
Hyundai Brake Disc,Hyundai Auto Brake Disc,Hyundai Car Brake Disc,Hyundai Automobile Brake Disc
Zhoushan Shenying Filter Manufacture Co., Ltd. , https://www.renkenfilter.com