Cover:
Chinese Association of Artificial Intelligence blockbuster release
"2018 Industry Innovation artificial intelligence assessment white paper"
" 2018 AI industry innovation assessment White Paper" by the Chinese Association of Artificial Intelligence, National Security Center work letter, the Chinese Academy of happiness industry, global security think syndication.
White Paper focused on artificial intelligence enabling technologies and application scenarios of the two levels, based on papers, patents, human resources, trade barriers multiple dimensions such as innovatively constructed artificial intelligence industrial innovation evaluation system, an objective assessment of the innovation in artificial intelligence industry level of development, the government, enterprises, investment institutions layout of artificial intelligence provides a reference and reference.
Scope of the study and evaluation system
with artificial intelligence to usher in the third wave of development in the world's major countries to actively promote, community investment and expectations of artificial intelligence has never been greater. Build a scientific and objective evaluation system innovation, to accurately assess the level of innovation and development in artificial intelligence industry, has important practical significance to the healthy and orderly development of artificial intelligence industry. Report in conjunction with the development and application of artificial intelligence technology level segment,
Focus speech interaction, text processing, computer vision and depth of learning four enabling technologies, as well as transportation, healthcare, manufacturing, security, retail and other eight key scenarios , the level of artificial intelligence to industrial innovation makes an objective evaluation.
An objective analysis of the four core enabling technology is currently in the development stage and eight key application scenario, a report based on innovative research to evaluate existing academic industry, combined with the industry attribute the industry of artificial intelligence, the use of quantitative and qualitative analysis, build a scientific and objective evaluation system of artificial intelligence industrial innovation. System under the enabling technology readiness index and application integration scenarios index level assessments two indexes and enable the establishment of technical readiness under the theory, application, performance-driven force three secondary indicators, the degree of integration scenarios under established resources, technology, data, scenarios, environmental driving force five secondary indicators.
Artificial intelligence enabling technology readiness
develop deep learning technology, to promote interactive voice, text processing, computer vision, artificial intelligence represented by the rapid development and rapid fall in multiple scenes. For the objective evaluation of the depth of learning, represented by four levels of development of enabling technologies, the report evaluated from theoretical research, applied research and technical performance of the three dimensions, respectively, to calculate the four enabling technology readiness index.
From enabling technology readiness index, the maximum depth learning readiness.
As mainstream artificial intelligence algorithms, deep learning Readiness highest ( 8.3 ), was in technical maturity ;
Computer Vision ( 7.7 ) and voice interaction ( 6.2 ), followed by the application of technology in the exploration stage, mainly in voice assistant medical diagnostic imaging and represented products have gradually entered the practical stage;
Text processing technology is still in the climbing stage, technical progress has been slow so there is a big distance from the real and practical.
From the actual development of enabling technology point of view, depth of learning and computer vision is the focus of the layout. In theory (paper output) connection, four enabling technologies from 2013 began to become a hot topic, in which deep learning is the focus of academic attention,
Followed by computer vision. Papers interactive voice and text processing of output growth is relatively stable, but the text processing throughput papers and citations are lowest.
In applied research (patent application) aspects,
Computer vision and depth of learning higher proportion of patent applications, patents but lower average intensity distribution patent still in its infancy;
Interactive voice low but higher proportion of patent applications mean intensity, indicating concern about the voice interaction showed a gradual downward trend.
From the energy level of the technical development of the two countries , the United States four theoretical and applied research in energy technology are significantly ahead of the Chinese.
In theoretical research, Sino-US gap in the field of text processing minimum, maximum gap depth of field of study;
In applied research, the smallest gap depth of field of study, the largest gap in the field of voice interaction.
Specifically, the United States four papers influence the average patent intensity and energy technology is much higher than China, Chinese papers and patents "and more but not strong" situation still exists.
We also found that China four enabling technologies patent application volume in the first place, especially related to higher R & D activity in recent years, more than 54 percent of all patent applications in the past three years.
From enabling the distribution of technical personnel , the United States four areas of artificial intelligence enabling technologies related to high-end talent far ahead of other countries.
Statistics found that US high-end artificial intelligence personnel over 13,000, China is less than 05,000, compared with the US gap.
From the sub-technical field of view,
High-end talent highest proportion of computer vision-related, up to 38%, of which the United States 5432 people, 1892 Chinese people.
Enabling R & D personnel from China distributed artificial intelligence point of view, Beijing, Guangdong, Jiangsu, Shanghai and Zhejiang provinces and five people have obvious advantages , including Beijing, Guangdong artificial intelligence research and development personnel more than 10,000 people.
Artificial Intelligence degree of integration scenarios
With the development of deep learning as the representative of enabling technologies, a large number of technology companies from specific industries or scene to promote artificial intelligence enabling technologies and industry to accelerate integration, to provide differentiated new products, services and solutions, to form rich "AI +" scenario has become an important driving force of the rapid development of artificial intelligence industry.
This report resources, data, and the scene environment five dimensions of the driving force of eight "AI +" scene evaluated, calculates the degree of integration of the eight scenarios.
Eight "AI +" scene: automotive, medical, furniture, retail, robotics, security, manufacturing, education
From the degree of integration scenarios index, artificial intelligence and the industry is still in its early integration of artificial intelligence. Depending on the application scenario fusion index display, automotive (3.9), medical care (3.8) and household (3.7) is a relatively high degree of integration of artificial intelligence three scenes; retail (3.5), robotics (3.3) and security (3.2) times of; manufacturing (3.0) and education (2.8) lower fusion index.
From the integration of the actual situation scenarios point of view, automotive, medical, home is the focus of the layout. In terms of technology the driving force, the artificial intelligence in all areas of patent applications since the beginning of 2014, explosive growth, including automotive and medical fields increased significantly, while education and relatively slow growth in the retail sector. The driving force in terms of resources, artificial intelligence research institutes and R & D personnel are mainly concentrated in the automotive, medical, home field, relatively few engaged in retail, education of artificial intelligence research and development institutions and personnel. The last three years combined patent application situation, the patent portfolio is focused mainly in the automotive, medical, home and security fields, and the integration of artificial intelligence robot is the new hot spot applications.
The overall level of global integration scenarios, the US application integration obvious advantages. In eight areas of application, the number of American AI researchers accounted for about half, and China in various fields of artificial intelligence researchers generally less than normal.
In terms of patent applications, in addition to the medical field, China's patent applications exceeded the size of the United States, have obvious advantages especially in the field of patents and two robot manufacturer.
In the patent application strength, the United States significantly ahead of China, Chinese patent quality to be improved. Specific to the application scenario, the United States Patent scale and intensity of artificial intelligence medical field significant advantages, Chinese manufacturing robots and artificial intelligence patent application strength has certain advantages.
From the main bottleneck in application integration scenario view. The lack of high quality data, high trade barriers, application scenarios is not clear is the current major bottleneck in artificial intelligence and depth of industry integration.
From the data the degree of accumulation of view, automotive, medical and robotics three data fields have a certain advantage, and home and create two areas of data accumulation obviously inadequate.
Data from the openness of view, a higher degree of openness three areas of automotive data, education and robotics, and medical and manufacturing data both areas is relatively low degree of openness.
Barriers to intervention from the scene point of view, the medical industry, higher barriers to manufacturing, artificial intelligence, enterprise less accessible.
Evaluation of artificial intelligence level of industrial development
through enabling technologies and application integration scenarios evaluated, we can see that the overall development of artificial intelligence is still in its infancy.
From enabling technology development, deep learning has become the mainstream of artificial intelligence algorithms, it is the key direction of theoretical research; deep learning technology has been in a mature and increasingly applied to various practical scenarios, also gradually emerging out of certain development bottlenecks;
Computer vision and voice interaction is still in the early application of technology, the two technologies have begun to try to apply landing in different scenarios;
The text is still in the process of technical climbing, slow technological progress. Fusion from the application point of view, based on the division of the report stage of development scenarios,
AI currently in incubation period integration in the automotive, medical, home, retail, robotics and security industry, while in its infancy manufacturing and education sectors still in the integration.
Industrial development of artificial intelligence judgments and Prospects
01 enabling technology for
voice interaction . Voice interactive technology exists for large-scale data dependent, low accuracy far-field recognition, poor complex scene recognition technology bottleneck effect , especially semantic understanding technology is not yet a real breakthrough seriously restrict large-scale commercial speech interactive technology.
The next step will focus on voice interaction to enhance the recognition rate in the far field identification especially complex environment, and smart home is still the best scene is undoubtedly the voice interactive technology to explore.
Text processing . Scene, learning and data acquisition are the major difficulties faced by text processing technology to enhance learning, visual language integration, joint study will be the next major breakthrough in the direction of text technology.
Text processing technology development will lead to a high level of digital penetration, low policy and social barriers, strong individual elements of the industry.
Computer Vision . Computer Vision bottleneck lies in the high degree of complexity, robustness is low, the lack of data and calculate power costs are too high.
Computer vision is focused on the development of the use of unsupervised learning and transfer learning ways to reduce data-dependent, field trial upgrade algorithm, and achieve integration with the depth of text, voice technology.
Deep learning . Deep learning depends on the gradient of multilayer neural networks fall under a number of parameters and the ensuing continuous optimization, but the result of a multi-layer gradient descent is non-linear and non-concave, the depth of the effectiveness of learning theory difficult to prove.
The future direction of the depth of learning is mainly draw on research to understand and model the actual depth learning mechanism.
02 scenarios Fusion
AI + car. Led to unmanned intelligent vehicle is a high degree of integration of artificial intelligence application scenarios, the traditional automotive industry will be the innovation of new technologies and business models. However, the development of intelligent vehicles are still faced with the vehicle hardware and software technology, artificial intelligence algorithms, and policy and commercialization of immature multiple challenges.
AI + Medical . The rapid development of intelligent medical field have appeared virtual assistant, medical assistance, intelligent imaging, drug development, precise variety of new medical practice. The number of underlying medical data quality is uneven, the lack of complex personnel system, the medical industry scenarios running difficult, high trade barriers and so restricts the depth of artificial intelligence applications.
AI + home . Integration of artificial intelligence and the home is the focus of current exploration industry. AI interaction, decision-making and service three levels of optimization to improve the performance of household products. Product price is high, it is difficult to protect user privacy, speech recognition is low, interoperability difficulties are the main challenges facing the development of the smart home.
AI + retail . AI booster online and offline retail industry, the depth of integration and bring further extend the consumer scene, enhancing the overall user experience of consumption. Currently, there are challenges based technology to enhance reliability and application scenarios. In addition, how to open up the B and the C-terminal end smart retail industry need to solve the problem.
AI + robot . AI pushing towards intelligent robot from mechanization. Intelligent robots in industry and services is becoming an important human assistants, such as assisting robots, robot logistics and public service robots. But the kinds of computer interaction, situational awareness and machine learning technology limitations, the current level of intelligence robot is still low.
AI + security . Artificial Intelligence applications in the security industry in the exploration stage. Intelligent security with algorithms, calculate force data as the three elements of development, the product is mainly reflected in the floor structure of the video, biometric identification, object recognition feature three aspects. Artificial intelligence will drive the security industry gradually to the city, integrated and proactive direction.
AI + manufacturing . Artificial intelligence from various aspects of research and innovation, quality control, fault diagnosis, operations management, promote the transformation and upgrading of the manufacturing sector, the driving force is the core of intelligent manufacturing. However, manufacturing and integration of artificial intelligence is still in cultivation period. The lack of high-quality industry data, lack of enterprise computing capabilities, communication standards coordination is not a major obstacle to achieve artificial intelligence and depth of integration of the manufacturing sector.
AI + education . Artificial intelligence technology used in the field of education, which can effectively improve teaching, learning, training, assignments, evaluation, management and other aspects, the rational allocation of teaching content Educational, Scientific implement individualized. The lack of high-quality learning trajectory data and the technology itself is not yet mature, resulting in artificial intelligence and the degree of integration in the field of education is far behind other industries.
-------- (I am dividing line) --------
reference:
1. Chinese Association of Artificial Intelligence newsletter http://www.caai.cn/index.php?s=/home/article/index/id/51.html
Remarks:
Initial modified: November 1, 2019 18:47:26
Environment: Windows 7 / Python 3.7.2
-------- (I am dividing line) --------
Note: This article is all summarized in the "China Association of Artificial Intelligence Communications", written primarily for personal blog and friends to learn information, and the spread of the original author's wisdom.
If readers want to see the latest information, you can enter the "References", download the appropriate information.
This data was downloaded from the "Society of Communication" in, if infringement, please inform my treatment.