Shulou(Shulou.com)06/02 Report--

Author | end of the cluster

Edit | Tokai

Standing at the end of 2019, this may be the last and biggest pulse of artificial intelligence in China this year.

On December 20, 2019, the two-day "New Generation artificial Intelligence Academician Summit Forum" sponsored by Pengcheng Laboratory and the technological Innovation Strategic Alliance of the New Generation artificial Intelligence Industry opened in Shenzhen.

The conference continued the strong lineup of academicians last year, and this year invited as many as 10 academicians and many top experts in the field of artificial intelligence and business circles. From their respective fields, the development status and future directions of artificial intelligence are analyzed in depth.

After the last time AI Science and Technology Review published several academicians Pu Muming, Gao Wen, Zhang Dongxiao, Zhao Qinping, Wu Jianping and industry experts Tang Xiaoou, Wang Haifeng, etc., they deeply felt the pulse of artificial intelligence.

This AI science and technology review will share the in-depth views on artificial intelligence of Zhang Zhengyou, Yan Shuicheng, Sun Jian, Chen Xilin, Li Shipeng and Xia Qin. At the same time, there are also five academicians.

Chen Jie, academician of the Chinese Academy of Engineering and president of Tongji University

Ding Wenhua, academician of the Chinese Academy of Engineering and radio and television technical expert.

Wang Shafei, Academician of Chinese Academy of Engineering and expert in Communication and Information system

Wang Lijun, academician of the Chinese Academy of Sciences and expert in laser and optoelectronic technology

Zhang Jianwei, Professor, Department of Informatics and Science, University of Hamburg, Germany and Academician of the Academy of Sciences in Hamburg, Germany.

Prospects and suggestions for the future development of artificial intelligence in China.

Let's take a look.

1. Zhang Zhengyou: the Intelligent Evolution of Robot

As the first guest in the afternoon, Dr. ACM and Dr. Zhang Zhengyou, Director of Tencent AI Lab & Robotics X, brought a report on the theme of "Intelligent Evolution of Robots".

Tencent

Director of AI Lab & Robotics X, ACM, IEEE Fellow Zhang Zhengyou

Before officially entering the report, Zhang Zhengyou recalled that last year, Academician Gao Wen also invited him to give a report at the Academician Summit Forum on artificial Intelligence, and when Academician Gao Wen invited him again this year, he also stressed that "the latest research report must be done." He joked with a smile: "I guess he wants to test me to see if I have made any research progress in the past year or so." "

"artificial intelligence is still in early spring, it is not very intelligent, and there are a lot of problems. After that, he pointed out that at present, artificial intelligence only learns from a large number of labeled data, and its generalization ability is poor, citing the fact that the camera cannot recognize the false image of the obscured lens.

He believes that with the development and full application of sensor technology, the era of coexistence of human and intelligent robots is bound to come, which is an important reason why he chose to return home to join Tencent to establish Robotics X Robotics Lab.

Dr. Zhang Zhengyou then introduced the six components of the robot, including ontology, perception, actuator, power system, interactive system and decision-making. The future trend of robots is automation, intelligence, and independent decision-making in an uncertain environment. Aiming at the autonomous decision-making of the robot, he proposed the SLAP paradigm, that is, sensors and actuators should be closely integrated to improve their ability and make decisions with the help of learning and planning modules.

With regard to the future breakthrough of intelligent robot technology, he once again mentioned the "A2G theory" shared last year, in which ABC represents the basic ability of the robot, A means that the robot can see, speak, listen and understand, B is the robot body, C is automatic control, while DEF refers to a higher level of robot ability, D is evolutionary learning, E is emotional understanding, F is flexible manipulation. The last layer, G, means to protect human beings. This puts forward requirements for more advanced and intelligent robots, and the ultimate goal of robots is to serve human beings.

Finally, Zhang Zhengyou expressed his vision for the development of robots, that is, man-machine coexistence, co-creation and win-win. To this end, it is necessary to create this future from four aspects: "using robots to enhance human intelligence, caring for human emotion, giving full play to human physical potential, and realizing man-machine cooperation".

2. Yan Shuicheng: Transform AI into Affordable Intelligence

Also present at the second artificial Intelligence Academician Summit Forum were Dr. IEEE and IAPR Fellow Yan Shuicheng, chief technology officers of Yitu Technology. In his speech on the theme "Transform AI into Affordable Intelligence", he pointed out the challenges of turning artificial intelligence into "Affordable Intelligence" and shared some work progress from the perspective of chips and models.

Yan Shuicheng, chief technology officer of Yitu Technology, IEEE and IAPR Fellow

Yan Shuicheng mentioned that the core mission of an AI startup and a large AI lab is to achieve the real landing of AI, which needs to solve two problems:

The first is the algorithm, on the one hand, to ensure that the algorithm is "usable", that is, the accuracy is high enough to really unlock a scene, and the other is that the algorithm should be "useful enough". Because now many scenarios based on single-mode algorithms can no longer provide satisfactory solutions for users.

The second is computing power, on the one hand, to make it "affordable" for users, such as the AI chip used to support computing, whose concurrent performance should be high enough; on the other hand, to make it "affordable" for users, that is, to ensure that the power consumption is low enough, otherwise, even if users buy home, they may not be able to use it because the electricity bill in the data center is too high.

As artificial intelligence is used in more and more scenarios, and as the technology has reached the stage where it can be used, it is not only more demanding for computing power and algorithms, but also more considered from the perspective of "Affordable".

Dr. Yan Shuicheng further pointed out that studies have shown that the computing power required to train and test artificial intelligence models doubles every three and a half months, which is much faster than Moore's Law. As at the NeurIPS meeting just held this year, one of the things that people pay more attention to is: as more and more AI applications are used, the electricity consumption is also getting higher and higher. Will this have an impact on the environment?

So to really land AI in a scene, the two most critical engines are:

The first engine is the high-performance AI model, which is the dimension of the algorithm. There are two ways to obtain a high-performance AI model, one is based on different Motivation patterns, and the other is based on NAS (Neural Network Architecture search) model. Aiming at this point, I mainly hope to solve the problem of "Affordable" in research and application.

The second engine is the high-performance AI chip, which is the dimension of computing. In view of this, chip manufacturers should first follow algorithms and chip principles to ensure that the chip achieves high performance in sufficiently large usage scenarios; secondly, it should predict the development trend of the most cutting-edge algorithms in the field to ensure that the chip can "give full play to its strengths" in the next few years; finally, the construction cost of users should be low enough and affordable.

Finally, he concluded that AI is used in more and more scenarios, and the pursuit of accuracy and goals are getting higher and higher, which puts forward higher and higher requirements for algorithms and computing power. At this time, the "Affordable" problem of AI will become more and more important.

Moreover, if we want to convert AI into "Affordable Intelligence", high-performance AI models and high-performance AI chips are the twin engines that drive this conversion, which is the only way to make our end users "affordable" and "affordable".

3. Sun Jian: the Frontier Development of Visual Computing

Dr. Sun Jian, the chief scientist and winner of the he Liang he Li Foundation Award, focuses on the research history and progress of computer vision from the convolution neural network and computer vision technology.

Sun Jian, chief scientist and winner of the he Liang he Li Foundation Award.

Convolution neural network started relatively early, and a professor in Japan put forward such a concept in the 1980s and was able to develop it. The follow-up research work on convolution neural network mainly focuses on four problems:

The first is the convolution of neural networks. Now what we use is 3 × 3 or 5 × 5 convolution, and convolution has experienced AlexNet network, GoogleNet network, faster R-CNN proposed by Facebook, ShuffleNet V1/V2 proposed by discerning technology, and so on. The latest research progress is dynamic convolution / conditional convolution.

The second is the depth of neural network. This is a problem that has plagued the neural network for many years. When the depth of the network is not large enough, it is difficult to realize the training of the network. The initial depth of the neural network is 8 layers, which increases to 20 layers after two years, and then the depth residual network proposed by Microsoft increases the depth to 152 layers. It can get good training results by using the idea of residual network.

The third is the width of the neural network. When the complexity of deep learning is more than one point, the larger the model, the error rate of training and testing will decrease at the same time, which is different from our traditional machine learning, which is actually related to the width of the network. At present, there are two relatively new directions: one is to start from the perspective of Kernel, and the other is to try pruning methods, such as MetaPruning.

The fourth is the size of the neural network. Generally speaking, the size of the neural network is constant in the process of training. however, the study found that better network performance can be achieved when the size of the neural network is changed in the training.

In view of the computer vision technology itself, Dr. Sun Jian focused on the direction of target detection and shared some problems and progress in the current research.

First, when the objects in the image are very close, the detection technology cannot accurately detect a single object.

Second, the design of computing architecture. In view of this, a lightweight two-stage target detector-ThunderNet is proposed, and the fusion of multi-scale architecture is designed, which runs very fast on ARM devices.

Finally, Sun Jian also pointed out some of the most important and most invested key issues in the application of computer vision:

First, data from special scenarios such as fires are very difficult to collect and difficult to obtain through data enhancement.

Second, the demand for new research methods such as self-supervision

Third, the problem of occlusion, although there is some progress in this regard, but deep learning can not completely solve this problem.

Fourth, deep learning and computer vision technology can not continuously track multiple objects that are dynamic at the same time.

Fifth, the problem of visual control, for example, it is not possible to continuously control the robot or manipulator through visual feedback.

Sixth, there are still great challenges to achieve low cost, easy deployment and security in practical applications.

Seventh, the existing methods can not achieve high-precision prediction.

4. Chen Xilin: towards understandable computer Vision

ACM, IEEE and IAPR Fellow Chen Xilin, researchers of the Institute of Computing of the Chinese Academy of Sciences, also brought a report on the theme of "towards understandable computer Vision" as guests of the report. In the report, he also shares some problems in the field of computer vision and his own exploration work on these problems from his perspective, and gives his own ideas on the future development of computer vision.

Researcher, Institute of Computing, Chinese Academy of Sciences, ACM, IEEE, IAPR Fellow Chen Xilin

He pointed out that it has been almost half a century since the concept was put forward, and it has mainly gone through four stages: Marr computational vision, active and objective vision, multi-view geometry and hierarchical 3D reconstruction, and learning-based vision. Although the progress made in this area is obvious, it also brings some problems, such as the emergence of evaluation benchmarks.

"before, people did not compare their studies with each other, even if they published papers, but the results may be at a standstill, so later there was a benchmark for evaluation, but one of the disadvantages is that today's researchers, especially students, only focus on" brushing the list ", which is not actually doing real research. So this is a big problem. "

He believes that in doing computer vision research, we need to know not only What and Where, but also How, Why, When and so on. In addition to the problems in research methods, the current computer vision research is also faced with two serious problems.

First, the research is in a "closed world", which shows that the new data can not be updated in time, can not borrow knowledge from other fields, and can not really understand the real relationship between objects.

Second, it is unable to deal with the problems of the open world well, for example, it is unable to distinguish between the language and semantics of the real world.

In order to solve these problems, Chen Xilin has made a series of explorations and work on interpretable decision-making model, similarity between conceptual space, semantic space and visual space, transferable comparative learning and the use of context.

Finally, he concluded that over the past 50 years, computer vision has achieved a lot of success in applications, so what will happen in the future? In the future, computer vision research will develop in an understandable direction, that is, the knowledge behind technology will play a more important role.

5. Li Shipeng: the Internet of everything is brilliant.

Li Shipeng, academician of the International Eurasian Academy of Sciences, vice president of Shenzhen Institute of artificial Intelligence and Robotics, and IEEE Fellow, focused on the development of IoT (Internet of things), AIoT (artificial Intelligent Internet of things) and IIoT (Intelligent Internet of things) in his speech entitled "Internet of everything, gathering talents".

Li Shipeng, academician of the International Eurasian Academy of Sciences, vice president of Shenzhen Institute of artificial Intelligence and Robotics, and IEEE Fellow

Li Shipeng believes that apart from other factors, the current era of artificial intelligence mainly includes four basic factors: AI, human, robot and IoT. Among them, human is the central factor, the interaction between human and intelligence is called man-machine coupling or man-machine cooperation, the combination of human and IoT is physical intelligence, and the combination of human and AI is a virtual scene.

The whole development process of IoT can be divided into three stages:

The first stage is the most basic stage of IoT. All the devices that can connect to the Internet and transmit data are called IoT devices, which mainly focus on the connection between devices, data collection and communication. People interact with IoT devices mainly through commands or remote control. At this stage, the degree of intelligence is very low, basically can only do IFDtt this type of conditional control.

The second stage is called AIoT, this term is not put forward internationally, it is a concept with Chinese characteristics. In the previous stage, IoT is basically unintelligent, and the application of data is very simple or only superficial, while in this stage, the data generated by IoT is intelligently processed. On the one hand, users' interaction with IoT devices becomes more and more intelligent; on the other hand, the collected data not only stay in the interpretation of the original data, but combine the data together to form some new knowledge. At this stage, AIoT always has a centralized controller to control all IoT devices, because it needs such a brain for overall control.

The third stage is IIoT (Internet of things). In the last stage, independent intelligent objects themselves have a certain degree of intelligence, and in many cases can operate independently. At this stage, we should explore how to combine the intelligence between intelligent and independent agents and what kind of intelligence can be formed. The relationship between man and machine has become a more equal cooperative relationship.

Li Shipeng believes that the evolution of aggregate intelligence brought about by IIoT is taking place, a trend that may break down some existing obstacles in the artificial intelligence industry and eventually pave the way for future AI frameworks based on causality.

6. Xia Qin: Yunda is beautiful-- the present and future of high-performance computing chips

Xia Qin, chief chip planner of Huawei Hayes and head of Haysturing's product management department, shared his thoughts on the present and future of high-performance computing chips under the title of "Yunda is Beautiful-- the present and Future of High-performance Computing chips". Taking Huawei's Teng chip and Kunpeng chip as examples, the problems of high-performance chip design are discussed.

Xia Qin, chief chip planner and head of product management department of Heysturing, Huawei

Xia Qin said that high-performance computing chips have become a very hot direction in the chip field, and many new theories and technologies have emerged in this field. Now engineers in this field are nothing more than solving a ternary equation, including three dimensions: performance, cost, and ease of use. If it is easy to solve from a single dimension, but if you want to achieve real mass production, you must solve the following three problems at the same time:

First, the problem of nuclear design can be explored in three directions: improving the main frequency, multi-core and micro-architecture (instruction set). For example, in the multi-core direction, Huawei has launched chips ranging from 8 cores to 16 cores to 64 cores. In terms of micro-architecture, Huawei's Kunpeng 920 chip has done some measures such as disordering and instruction prefetching to improve performance, and good results have been obtained.

Second, the problem of on-chip design, Huawei's optimization work on the chip includes using the latest technology to reduce the chip area, improve the yield, and then reduce the cost.

Third, the problem of peripheral interface, in this aspect, Huawei has made a lot of innovations in the memory channel interface of the chip, including interface rate, multi-IP interconnection and accelerator.

In addition, the design of AI chip also faces unprecedented challenges in software: first, AI-based computing requires more parallel processing power; secondly, the great difference between AI chip and CPU chip is that the former is truly heterogeneous computing. Finally, the construction of AI software stack is also faced with great challenges, such as the development process from Huawei Kunpeng software stack to Pengteng software stack, which is very complex and difficult.

She further summed up the future trend of chip design:

First, the performance of single thread and single core will continue to improve in the future, and there will be more definitions for instruction set. in addition, with the vigorous development of ARM, it will also provide a lot of room for future user customization, thus greatly improving the overall performance of the chip.

Second, although there is no good technology in memory storage, there is still a lot of room for exploration in the redefinition of memory interface and the whole memory architecture.

Third, new standards and interface technology breakthroughs, there will be new technological attempts in this area, but the results will be slower.

Fourth, flexible power management, with the arrival of heterogeneous computing in the future, low-load, low-power tuning technology will become very important.

"AI has many directions that can be developed in the future, and it is not a single-dimensional development direction. Although we do not know when the real AI era will come, I think computing power, collaboration and application are the three key dimensions that the whole AI technology can move towards full commercial use in the future. "

7. Five Academicians discuss the Future of AI

The second round table of academicians was held in the afternoon and was presided over by Chen Jie, academician of the Chinese Academy of Engineering and president of Tongji University. Ding Wenhua, academician of the Chinese Academy of Engineering and expert in radio and television technology, Wang Shafei, academician of the Chinese Academy of Engineering and expert in communications and information systems, Wang Lijun, academician of the Chinese Academy of Sciences and expert in laser and optoelectronic technology, and Zhang Jianwei, professor of informatics and science at the University of Hamburg in Germany and academician of the German Academy of Sciences in Hamburg, expressed their prospects and suggestions on the future development of artificial intelligence in China.

Academician Ding Wenhua: this conference is rich in content, which shows that China has made great progress in artificial intelligence in various fields. This year's speech covers algorithms, applications and resources, and everyone has made breakthroughs in their respective research directions. I believe that through the platform of Pengcheng Laboratory, we can gather high-end experts and talents in the field of artificial intelligence from all over the country and the world to communicate and promote the development of artificial intelligence technology as a whole.

Academician Wang Shafei: I think the whole development of artificial intelligence still needs to go through a long stage. There are several challenges and problems.

First, when the technology of artificial intelligence is applied to the ground, there are still many problems in intelligent reasoning, and it is difficult for artificial intelligence to reason unknown scenes or targets like human beings.

Second, the interpretable problem. Now AI can calculate a large amount of big data and achieve certain perception, but is the result correct? Maybe we can improve this problem by adding human experience in the future.

It is very enlightening to listen to the report of the experts today. I think that through the efforts of my colleagues, artificial intelligence can break through the difficulties in basic research and get better application.

Academician Wang Lijun: I mainly study laser chips. In recent years, with the development of artificial intelligence and information perception, I have also carried out research on communication and information perception (optoelectronic integrated chips). I have studied laser chips for decades and have some experience of my own.

The development of domestic chips has been relatively slow in recent years, and it is still in the stage of being limited by people. Why is there such a situation? There are several reasons:

First, the cost of equipment needed in the process of chip development is so huge that ordinary units cannot afford it.

Second, in terms of time, it takes several years or even more than ten years to make a chip.

Third, in recent years, we have been pursuing to produce results as soon as possible, of course, the intention is good, but some things should respect the facts, like this kind of large investment, slow results, some government agencies may not be very willing to vote.

Fourth, chip researchers, especially young people, are also more willing to invest in research that can achieve quick results and produce articles immediately.

Now our country is aware of these problems and has taken some measures to overcome the chip problem. I believe that in a few years, there will be a major breakthrough in the chip in our country.

In addition, for information perception, I personally think that the next step is a very important direction in optoelectronic and hybrid integrated chips, which not only further integrates integrated circuit technology and integrated optics technology, but also integrates perceptual software and optical things together, further improving reliability, and will greatly promote the AR industry.

Academician Zhang Jianwei: first of all, we warmly congratulate Pengcheng Laboratory on its great achievements in more than a year. After defining several important directions in the future, we have organized researchers from all walks of life to solve major international livelihood issues from an intersecting point of view. it has become a very important platform for attracting the highest-level talents of industry, university and research to gather, communicate and collide with each other.

After listening to your reports today, I would like to emphasize a few more points:

First, lay the foundation. Today, some experts talk about multimodal technology, from brain science multimodal processing, chip multimodal processing, image recognition, picture understanding and so on, multimodal information processing has become a core technology of artificial intelligence, and it is also worthy of our further development and research. The cross-modal learning project I organized five years ago is the largest research project between China and Germany, which organizes the study of human multimodal learning mechanism from the aspects of brain science, psychology, artificial intelligence, robot, and so on, and then makes a new algorithm. finally, it is realized by a robot.

Second, how to land artificial intelligence next. In my opinion, in addition to providing a basic platform for artificial intelligence, the next step is to really integrate the demand and vertical fields, deeply integrate, deepen and lengthen the processing chain, and create a world-class intellectual property rights and a world-class market, so that the value of artificial intelligence can be generated faster.

Third, public platform communication, ecological building and social impact. Now the local government is also paying more and more attention to open source and providing a platform for enterprises. I think a new world-class artificial intelligence Demo prototype can be made in Shenzhen.

Https://www.toutiao.com/i6773753853239099918/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.