Artificial Intelligence in China - A Snapshot from the Chinese Web
3 likes514 views
The report analyzes Chinese online discussions about artificial intelligence from December 2018 to January 2019, highlighting China's aim to lead in AI development amidst social and ethical challenges. Key areas of focus include education, career opportunities, and the dominance of machine learning technologies, particularly neural networks. Major AI players in China like Baidu, Alibaba, and Tencent are mentioned, along with emerging companies and educational initiatives in the field.
![Popular EV/PHEV modelssResearch and education institutions
As seen on the previous slide, education and the
supply of AI talent are a central topic and a critical
component in the implementation of a successful AI
strategy. According to McKinsey¹, China needs to
build three layers of talent:
● Top scientists pushing the boundaries of
fundamental AI technology
● Developers who can create AI applications
for real-world contexts
● Large base of workers who work alongside
AI systems on a day-to-day basis
The chart shows the distribution of Chinese
universities in the Web data. Qinghua University
clearly stands out - the school launched an AI
institute in July 2018 and hired Google’s AI lead
Jeff Dean as advisor.² Beijing university also
announced a new campus with focus on AI. In total,
34 Chinese universities currently have specialized
AI institutes.³References:
[1] McKinsey. 2017. Artificial Intelligence: Implications for China.Retrieved from https://mck.co/2Ri33ag.
[2] Qinghua University. 2018. 清华与谷歌联合举办人工智能学术研讨会. Retrieved from https://bit.ly/2RHuD0b.
[3] 程瑶. 2018. 北大将在昌平建新校区以人工智能为特色. Retrieved from
http://www.xinhuanet.com/local/2018-11/12/c_1123696834.htm.
Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 3](https://image.slidesharecdn.com/aiinchina-2-190114110800/85/Artificial-Intelligence-in-China-A-Snapshot-from-the-Chinese-Web-4-320.jpg)
![Popular EV/PHEV modelssZooming in on the Neural Network
References:
[1] Yoav Goldberg. 2016. A primer on neural network models for natural language processing. J. Artif. Int. Res. 57, 1 (September 2016), 345-420.
[2] Alibaba. 2019. 达摩院2019十大科技趋势. Retrieved from https://www.iyiou.com/p/89119.html.
[3] Ian Goodfellow et al. 2014. Generative adversarial nets. In Proceedings of the 27th NIPS.
Neural Networks come in different architectures,
depending on the structure of the layers and the way in
which data travels through the network.¹ The chart
illustrates the distribution of different network types in
the data.
Graph Neural Networks (GNNs) have been around for
some years; interest was revived in the past months by
Chinese scientists. Alibaba names the GNN as one of
the main research trends for 2019.² Convolutional and
Recurrent Neural Networks, incl. the highly efficient
LSTM architecture, are more traditional architectures
and widely used in NLP and image recognition.
Finally, the Generative Adversarial Network (GAN) is a
relatively new architecture proposed by the university of
Montreal.³ GANs detect, learn and reproduce the natural
features of a data set. By virtue of their generative
capacity, GANs have been especially intriguing for
creative tasks such as the creation of paintings and
videos, and thus are an important milestone in moving
the AI frontier.
Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 5](https://image.slidesharecdn.com/aiinchina-2-190114110800/85/Artificial-Intelligence-in-China-A-Snapshot-from-the-Chinese-Web-6-320.jpg)
![Popular EV/PHEV modelssAI players beyond BAT
This chart shows prominent AI players beyond BAT and
classifies them by application.
Yixue Education has generated a lot of discussion in the
past months with provoking hypotheses about the
interplay between AI and human creativity. In December
2018, Yixue started an experiment in Weifang
(Shandong), during which most schools in the city
implement the company’s adaptive AI-based learning
technology of the company.
Megvii is the owner of the widely used face recognition
platform Face++, which is also used by the Chinese
government and police. The company is considering an
IPO on the Hong Kong Stock Exchange in 2019.
UBTECH, the third player in the ranking, is B2C robotics
startup that secured USD 280M funding in May 2018.
References:
[1] 奎文新闻. 2018. 全市首批乂学教育松鼠AI实验基地在奎文区试点启动. Retrieved from https://kuaibao.qq.com/s/20181225B0GDDF00?refer=cp_1026.
[2] Bloomberg. 2019. Alibaba-Backed AI Startup Said to Mull Up to $1 Billion IPO. Retrieved from https://bloom.bg/2ClPvAo.
[3] Ian Goodfellow et al. 2014. Generative adversarial nets. In Proceedings of the 27th NIPS.
Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 7](https://image.slidesharecdn.com/aiinchina-2-190114110800/85/Artificial-Intelligence-in-China-A-Snapshot-from-the-Chinese-Web-8-320.jpg)
Artificial Intelligence in China - A Snapshot from the Chinese Web
- 1. Artificial Intelligence in China Competitors Markets Consumers Industries Partners A quantitative snapshot of common topics from the Chinese Web (Dec 2018 - Jan 2019)
- 2. Popular EV/PHEV modelssOverview China plans to become a leading player in AI in the next few decades - this ambition, together with numerous social and ethical challenges, turns Chinese AI into a highly vibrant, but also controversial topic. The present report analyzes Chinese online texts pertaining to AI in the period of December 10, 2018 to January 5, 2019. The data is drawn from Anacode’s Social Technology Index, which includes a variety of data such as social media, business news and specialized technology portals.The analysis is based on a fine-grained representation of the AI domain which, among others, models the major technologies, players and application areas. For more information on Anacode’s technology, please download our whitepaper. Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 1
- 3. Popular EV/PHEV modelssArtificial Intelligence in context This chart shows the various contexts in which AI is discussed on the Chinese Web. The most prominent area is Careers & Education. This can be explained two tendencies: on the one hand, education is one of the main application areas of AI; city-wide experiments in AI-based education turn it into a hot discussion topic (cf. slides 6 and 7). On the other hand, China is currently making huge efforts to systematize and scale training and education in AI itself. This is reflected in numerous online tutorials, course descriptions and reviews of the various offerings. Naturally, AI is also an attractive career path for engineering and computer science graduates. Beyond education and HR, infrastructural areas relating to Smart Cities, Finance and Energy make up another large proportion of the data. Pure B2C applications such as Games and Home Appliances make up a relatively small proportion of the data. Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 2
- 4. Popular EV/PHEV modelssResearch and education institutions As seen on the previous slide, education and the supply of AI talent are a central topic and a critical component in the implementation of a successful AI strategy. According to McKinsey¹, China needs to build three layers of talent: ● Top scientists pushing the boundaries of fundamental AI technology ● Developers who can create AI applications for real-world contexts ● Large base of workers who work alongside AI systems on a day-to-day basis The chart shows the distribution of Chinese universities in the Web data. Qinghua University clearly stands out - the school launched an AI institute in July 2018 and hired Google’s AI lead Jeff Dean as advisor.² Beijing university also announced a new campus with focus on AI. In total, 34 Chinese universities currently have specialized AI institutes.³References: [1] McKinsey. 2017. Artificial Intelligence: Implications for China.Retrieved from https://mck.co/2Ri33ag. [2] Qinghua University. 2018. 清华与谷歌联合举办人工智能学术研讨会. Retrieved from https://bit.ly/2RHuD0b. [3] 程瑶. 2018. 北大将在昌平建新校区以人工智能为特色. Retrieved from http://www.xinhuanet.com/local/2018-11/12/c_1123696834.htm. Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 3
- 5. Popular EV/PHEV modelssFundamental technologies and algorithms From a technological perspective, there are two main approaches to AI: - Machine Learning: the machine learns from large quantities of data. - Rule-based approach: a human expert teaches the machine by coding rules on a specific domain. The chart shows that Machine Learning with its subdomain Deep Learning and the Neural Network as its principal tool stand out and clearly win over rule-based approaches such as Expert Systems. Traditional “shallow” learning algorithms, such as Support Vector Machines and different types of regression, only make up a small proportion of the data. While still being widely used in practical applications, these methods attract less interest in the research and coding community and thus … . Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 4
- 6. Popular EV/PHEV modelssZooming in on the Neural Network References: [1] Yoav Goldberg. 2016. A primer on neural network models for natural language processing. J. Artif. Int. Res. 57, 1 (September 2016), 345-420. [2] Alibaba. 2019. 达摩院2019十大科技趋势. Retrieved from https://www.iyiou.com/p/89119.html. [3] Ian Goodfellow et al. 2014. Generative adversarial nets. In Proceedings of the 27th NIPS. Neural Networks come in different architectures, depending on the structure of the layers and the way in which data travels through the network.¹ The chart illustrates the distribution of different network types in the data. Graph Neural Networks (GNNs) have been around for some years; interest was revived in the past months by Chinese scientists. Alibaba names the GNN as one of the main research trends for 2019.² Convolutional and Recurrent Neural Networks, incl. the highly efficient LSTM architecture, are more traditional architectures and widely used in NLP and image recognition. Finally, the Generative Adversarial Network (GAN) is a relatively new architecture proposed by the university of Montreal.³ GANs detect, learn and reproduce the natural features of a data set. By virtue of their generative capacity, GANs have been especially intriguing for creative tasks such as the creation of paintings and videos, and thus are an important milestone in moving the AI frontier. Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 5
- 7. Popular EV/PHEV modelssThe BAT race Baidu, Alibaba and Tencent are the core players in China’s AI arena. By virtue of their resources, these companies can tap into the combined potential of algorithmic research, huge data quantities and powerful hardware and computation resources. The following chart compares the top 5 application areas of Baidu with the offerings of Alibaba and Tencent: Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 6
- 8. Popular EV/PHEV modelssAI players beyond BAT This chart shows prominent AI players beyond BAT and classifies them by application. Yixue Education has generated a lot of discussion in the past months with provoking hypotheses about the interplay between AI and human creativity. In December 2018, Yixue started an experiment in Weifang (Shandong), during which most schools in the city implement the company’s adaptive AI-based learning technology of the company. Megvii is the owner of the widely used face recognition platform Face++, which is also used by the Chinese government and police. The company is considering an IPO on the Hong Kong Stock Exchange in 2019. UBTECH, the third player in the ranking, is B2C robotics startup that secured USD 280M funding in May 2018. References: [1] 奎文新闻. 2018. 全市首批乂学教育松鼠AI实验基地在奎文区试点启动. Retrieved from https://kuaibao.qq.com/s/20181225B0GDDF00?refer=cp_1026. [2] Bloomberg. 2019. Alibaba-Backed AI Startup Said to Mull Up to $1 Billion IPO. Retrieved from https://bloom.bg/2ClPvAo. [3] Ian Goodfellow et al. 2014. Generative adversarial nets. In Proceedings of the 27th NIPS. Data basis: Web data from Anacode China Social Technology Index, Dec 2018 - Jan 2019. 7
- 9. Popular EV/PHEV modelssContact Anacode GmbH Kurfürstendamm 76 10709 Berlin info@anacode.de www.anacode.de Dr. Janna Lipenkova +49 (152) 098 17 228