The world professional Go ranking website (Goratings) has just released the world Go ranking, with Chinese player Ke Jie ranked first in the world with a score of 3647. However, before this top Go master, there is an opponent with nearly perfect strength that is difficult for humans to defeat in the short term, and that is the constantly upgraded Google AlphaGo.
AlphaGo has continuously defeated top human players, triggering a new wave of artificial intelligence development. Internet giants like Google, Baidu, Alibaba, Tencent, and Facebook have all invested heavily in the field of artificial intelligence (AI), especially in AI, cloud computing, and big data, which form a triangular support relationship in industrial logic. It seems that the whole world is trying to fully cloudify AI, using cloud-based AI as an important driving force for fueling the digital economy and promoting digital transformation.
However, at MWC 2017 Shanghai, we saw many embedded AI solutions from Zhongke Chuangda. Why did Zhongke Chuangda choose the development path of embedded AI? What are the differences in technical principles and application scenarios between embedded AI and cloud AI? What is the industrial value of embedded AI? With these questions in mind, “Lao Shang Looks at Technology” interviewed Mr. Yang Yuxin, Vice President of Zhongke Chuangda.
Most people are doing cloud AI; why did Zhongke Chuangda choose embedded AI?
The trend of AI becoming resource-based, socialized, and service-oriented is becoming increasingly clear. Whether it is ET mentioned at the Alibaba Cloud Summit in Shanghai or the so-called smart cloud at the 2017 Tencent Cloud + Future Summit, the essence is to turn AI into a public service resource like tap water, electricity, and natural gas. Users can easily access cloud AI by turning on the valve or switch. Currently, whether it is Google, Amazon, or China’s BAT, the AI implementation method they choose is mainly cloud AI, where users’ AI needs require network connections to data centers for responses. The biggest feature of embedded AI, however, is the localization of AI, which can free itself from the constraints of the network.
“In fact, cloud AI and embedded AI are not opposing relationships; they are more of a complementary relationship,” Yang Yuxin defined the relationship between embedded AI and cloud AI in an interview. Yang Yuxin stated that cloud AI is supported by the powerful computing power of data centers, making it more suitable for the early exploration of machine learning algorithms. As machine learning algorithms gradually take shape and mature, these algorithms can be placed into embedded AI systems to obtain AI services locally.
Yang Yuxin made a vivid analogy about the different implementation paths of AI: cloud AI is like a tap water supply system built through water pipelines, providing a large number of AI services that can be used at any time; however, the tap water supply network does not cover every corner. For example, in the wild or on the road, if we have a need for water, a bottle of mineral water can solve the problem. Embedded AI is like bottled mineral water, solving the user’s need for AI services that can be accessed anytime and anywhere.
When talking about Zhongke Chuangda’s initial intention to choose embedded AI, Yang Yuxin said: “Zhongke Chuangda’s main business in the past was to develop embedded intelligent operating systems, with many years of R&D investment and experience accumulation in intelligent operating system technology, and a deep accumulation in artificial intelligence algorithms. We have close cooperation with upstream chip manufacturers. Zhongke Chuangda’s embedded artificial intelligence platform provides a complete set of solutions from the chip layer, driver layer, operating system layer to algorithm layer, aimed at enhancing the local real-time environmental perception, human-computer interaction, and decision-making control capabilities of intelligent terminal devices, rather than having intelligent devices become inactive without network support.”
In the era of the Internet of Everything, how does embedded AI demonstrate its unique value?
The commercial use of 5G may be realized in the next three to four years. In the era of the Internet of Everything, where can the value of embedded AI be fully realized in various application scenarios?
Yang Yuxin stated that if the Internet of Everything is truly realized, a derivative problem will arise, which is the explosion of data. According to research institutions like IDC, the global data volume is expected to grow at an annual rate of 50% in the coming years, and by 2020, the total global data volume could reach 40ZB. It is estimated that China’s data volume in 2020 will be 8.6ZB. Data statistics show that in 2015, China’s mobile data volume was about 400TB. Although the existing network infrastructure will continue to upgrade, it will still face a very real bottleneck in network bandwidth in the face of the data explosion brought about by the Internet of Everything. Therefore, gathering all data to the cloud, processing it there, and then distributing it to intelligent devices may not be the optimal choice.
For example, surveillance cameras generate a large amount of video data, but many images remain unchanged for a long time. If the camera is equipped with embedded AI, it can deduplicate at the source, retaining only valuable footage based on algorithms, achieving data cleaning and deduplication locally, which can greatly save storage and transmission costs. This scenario can fully demonstrate the commercial value of embedded AI.
Caption: Smart driving simulation experience
In vehicle systems, the value of embedded AI is also very obvious. During high-speed driving, whether it is autonomous driving or assisted driving, it requires millisecond-level responses to changes in road conditions. If purely relying on cloud AI, perhaps the 5G network can meet the communication latency requirements between vehicles and cloud AI under normal conditions. However, once entering tunnels or mountainous areas with poor signal coverage, high-speed vehicles may experience response delays. Therefore, in vehicular applications, embedded AI currently holds irreplaceable value.
In Yang Yuxin’s view, both cloud AI and embedded AI have their strengths and weaknesses. In reality, whether in mobile application scenarios like autonomous vehicles or assisted driving, or in smart home applications like ECHO, the ideal and reliable approach is to have cloud AI and embedded AI work together. Therefore, Zhongke Chuangda’s future focus will be on industrial collaboration and integration innovation with partners.
The future wars will be ecological wars; how to build an embedded AI ecosystem?
The development of the technology industry is changing rapidly, and the past dominance of tech giants will become history. Future competition in the technology industry will increasingly be based on ecosystems, and embedded AI also faces the task of building an ecosystem to achieve future ecological positioning.
According to Yang Yuxin, Zhongke Chuangda has now launched a development platform for smart hardware cost – the TurboX Intelligent Brain platform, aimed at leveraging an open ecosystem to gather industrial collaborative innovation forces and accelerate the process from smart hardware prototypes to productization. This intelligent brain platform integrates resources from the industrial chain, including content, applications, cloud services, etc. In the fields of drones, robots, virtual reality, smart cameras, wearable devices, and many other smart hardware, many star products, such as Weijun VR, Zero Degree Control DOBBY drone, Zero Zero Infinity HoverCamera, and Cat Eye Vision “Cat Mirror”, are all equipped with the TurboX Intelligent Brain platform.
Caption: A subsidiary of Zhongke Chuangda, ThunderView, demonstrated food recognition, able to identify food names, weights, and calories in real-time without an internet connection.
Yang Yuxin also stated that Zhongke Chuangda currently has deep cooperation with companies like ARM, Qualcomm, Intel, and Samsung. On one hand, we have jointly operated multiple laboratories with Qualcomm, Intel, and Microsoft, providing advanced resources and technical support for component adaptation and terminal testing; on the other hand, we have established joint ventures with ARM and Qualcomm – Anchuang Space Accelerator and Chongqing Chuangtong Lianda, to provide industrial platforms for smart hardware companies and innovators, promoting the development of smart hardware technology. At the same time, we have jointly established industrial investment funds with Lin Kong Tou, Angel Fund, and Polar Light to integrate resources to support innovation. In 2016, Zhongke Chuangda acquired the smart in-vehicle interaction technology company Rightware for 64 million euros, aiming to strengthen industrial collaboration or integration innovation in the embedded AI ecosystem.
According to relevant research institutions, the global artificial intelligence industry scale is expected to exceed 100 billion dollars in the coming years. However, I believe that the greater value of artificial intelligence will help promote many traditional industries to upgrade and evolve, such as intelligent medical care that can predict human lifespan and malignant diseases, intelligent transportation that can greatly reduce the probability of traffic accidents, and intelligent manufacturing that can better meet personalized needs… AI is playing the role of a new technological engine, accelerating the process of intelligentization of human society. In my view, cloud AI and embedded AI are like two legs working together; they complement each other to enable AI to be more widely and rapidly adopted. We have reason to expect that with the establishment of more and more open AI ecosystems, the somewhat sci-fi intelligent society is becoming increasingly within reach.