With the acceleration of urban digitalization, how does AI overcome the problem of long-tail scenarios?

In most sci-fi films, the forms of future cities have several commonalities: people can interact with all public facilities, visualization of public transportation data, real-time data transmission over long distances between people and people, and highly intelligent community and building control.

The ultimate goal of smart cities is also the same. Everything, including people, can be freely interconnected, all information is transmitted using data as a carrier, and any problems and needs can be solved through data collection and analysis. According to the above settings, the current smart city development is still at a very early stage, but the society’s pursuit of urban digitalization has been accelerating.

At the beginning of this year, Shanghai officially released the “Opinions on Comprehensively Promoting the Digital Transformation of Shanghai City”. This year’s Shanghai government work report also pointed out that in 2021, Shanghai will promote digital industrialization and industrial digitization, and accelerate the development of a new online economy. According to the “White Paper on the Development of China’s Digital Economy (2021)”, the scale of the domestic digital economy was 39.2 trillion yuan last year, accounting for 38.6% of GDP, and the growth rate was more than three times faster than GDP. There is no doubt that a thriving digital economy has become the key to driving sustained and long-term economic growth.

Now, urban digitization is gradually moving from point and line to surface dimension, showing stronger integration, and the direct effect of integration is the exponential improvement of operational and governance efficiency requirements, as well as the general adaptation of scenarios. In fact, this is still a matter of AI technology and layout, such as the amount of computing power, the precision of the algorithm, the fault tolerance rate and scheduling ability of the platform, etc. These projects involve long-tailed urban governance scenarios such as bicycle parking and garbage exposure, which often require all-weather and high-precision management and control. Therefore, the AI ​​systems of participating companies will also be affected in terms of visual recognition, rapid judgment, and timely response capabilities. put forward more stringent requirements. Urban governance is eager for “high-speed efficiency” According to the National Bureau of Statistics, the urbanization rate of my country’s permanent population has exceeded 60% in 2020.

Population increase and agglomeration have brought greater pressure on urban governance. Coupled with some governance problems brought about by the development of the new economic model based on sharing, the difficulty of urban governance is actually increasing. There are also some typical livelihood problems, such as high-altitude parabola, motorcycles entering elevators, and reasonable garbage classification. The superposition of these long-tail scenarios also brings new challenges to urban governance, especially in governance efficiency and resources.

Traditional digital governance methods have been difficult to adapt to these complex long-tail scenarios. On the one hand, traditional methods are mainly based on simple dataization, and do not form a complete closed-loop logic and solution with offline governance. On the other hand, because traditional methods are intelligent The degree is low and cannot meet the requirements of large-scale and multi-scenario. For example, in traffic travel, traditional visual solutions can only store, record, photograph, and identify common behaviors that need to be managed and controlled, but cannot judge and respond to abnormal events such as traffic jams and random parking. On-site management.

Another example is the focus of social and livelihood issues such as motorcycles entering elevators. Traditional visual solutions often do not have the ability to recognize. Even the intelligent solutions given by some companies have obvious shortcomings in recognition accuracy and post-recognition response. . Complicated long-tail scenarios and high-frequency visual analysis, from the perspective of AI technology, are actually a matter of computing power and algorithms. With a large scale and more requirements, the computing power must be strong enough to process and analyze these huge dynamics. The data and algorithms must also be matched, so that they can be analyzed and processed quickly and accurately. In general, as the number of scenes accommodated by the objective environment increases, the differences between the scenes increase, and the difficulty of urban digital governance increases accordingly, which puts forward higher requirements for efficiency and accuracy.

Hard solution of AI ecology Since the objective requirements of urban digital governance have become higher, only stronger AI solutions can be adapted accordingly. In fact, there are also two problem-solving ideas here. One is to overcome single scenarios one by one, but the ultimate unified management is realized. The difficulty is high. The other is to establish a basic solution first, and then superimpose and adjust it according to different scenarios, which requires the foundation to be hard enough. At the 2021 World Artificial Intelligence Conference, SenseTime released a blockbuster product called “SenseCore AI Big Device”, which is positioned as an AI infrastructure that can solve enterprise services, urban management, personal All kinds of long-tail application problems of life.

With the acceleration of urban digitalization, how does AI overcome the problem of long-tail scenarios?

Although it is infrastructure, this big AI device is divided into three layers of computing power, platform, and algorithm, and each layer has different functions and capabilities. For example, in terms of computing power, SenseTime has established Asia’s largest artificial intelligence computing power center in Shanghai. In terms of algorithms, it provides enterprises with personalized algorithm tools based on the model output from the platform layer. So far, more than 17,000 algorithm models have been developed.

The foundation of strong computing power, coupled with the continuous iterative algorithm blessing, is also a relatively unified development idea for the AI ​​and cloud computing tracks, but the differentiated competitiveness of each product is precisely because of this. Without platform-based collaborative integration, the overall efficiency, flexibility, and cost will be greatly reduced.

As far as the specific application is concerned, the core value of SenseTime’s large-scale device logic is that it can output different and useful solutions based on the same system. For example, Ark City Development Platform, in which SenseFoundry Traffic (Shangtang Ruitu) covers the three scenarios of rail traffic, traffic management and high-speed, and can make intelligent identification and rapid decision-making according to the needs of different scenarios; The solution, through the integration of perception, vision, and data capabilities, can perform digital and intelligent empowerment for scenarios such as parks, production, and emergency.

The difficulty of urban digitization lies not only in the scale of the scene, but also in the dimension of the scene. For example, compared with the individual level, enterprise-level digitization often requires the ability to penetrate the entire industry chain or the entire scene. Therefore, on the basis of a set of systems, to Effectively covering all scenarios, obviously requires the system to have strong compatibility and classification processing capabilities, as well as platform-based functional attributes.

The solution to the long tail problem For city managers, solving all problems with one system is the ultimate perfect solution, because it has high efficiency, low cost and good effect. Especially in various long-tail scenarios that are currently more prominent, if there is a system that can manage these scenarios in a unified manner, the governance efficiency will be amazingly improved. In fact, the goal of AI industry development is also the same. AI+ enterprises and industries are a set of logic to meet all needs and solve all problems. Now we have seen many unified enabling products in the fields of industry and agriculture, which is also the continuous fermentation of this trend. Where is the advantage of such a methodology, the application results of SenseCore AI big device can give some reference.

During the pilot period of “AI + One-Network Unified Management” in Changning District, Shanghai, more than half of the incidents were handled within 4 hours, and the fastest only took 20 minutes. In elevator-related scenarios, the escalator safety intelligent response system jointly built by SenseTime and Schindler last year can dissuade unsafe behaviors by voice at the entrance and issue early warnings at the exit based on the density of people; the wisdom jointly created by SenseTime and Evergrande The community can effectively identify unsafe behaviors such as motorcycles entering elevators and high-altitude parabolic objects. It can be seen that with the help of AI, the processing of long-tail scene problems has been significantly improved in two aspects, one is higher efficiency, and the other is a wider processing range. The existence of these long-tail scenarios is characterized by high repetition, that is, the processing logic and needs of different communities and cities are the same, and the processing of single or regional scenarios can obviously be effectively and quickly replicated and promoted. However, it should be noted that although the processing logic of the long-tail scenario problem is the same, the massive data that can be superimposed will also bring certain challenges to the computing power and algorithms.

In the long run, the core of the long-tail scenario problem is actually the efficiency and accuracy of AI judgment. Although the occurrence of critical point behaviors in the scene may be low-frequency, AI must conduct high-frequency analysis and judgment in order to prevent small and gradual progress. Not a patchwork. Smart cities are inseparable from a high-quality AI base. More cities accelerate the objective background of digital reform. In fact, it is the improvement of AI industrialization capabilities. In the early years, AI technology was still code. The process from 0 to 1 did not depend on the application depth and breadth, but now it is different. The co-evolution of AI, big data, cloud computing, Internet of Things, 5G and other technologies is giving AI applications. Open a superspace.

AI is a new infrastructure, just like the Internet in the past, it will eventually be added to all industries, but it is much more valuable than the Internet. AI can become the strongest productivity factor, driving the upgrading of the industry’s production methods and production logic, and ultimately the unit Produce more and better.

The SenseCore AI installation is such an infrastructure facility. For cities, its significance is not only to provide better digital solutions in the three areas of governance, industry and life, but also to provide cities with a comprehensive digitalization and life. Intelligence has created an extremely solid and inclusive urban base. On this urban base, seamless and efficient interconnection of all things will be easier and faster to achieve. But no matter how powerful this infrastructure is, AI must not cross the moral and ethical boundaries. Inclusiveness is the ultimate goal pursued by all technologies, but technology always serves people and cannot put the cart before the horse. Therefore, AI ethical governance is to identify and master this degree determined by people.

At WAIC 2021, Xu Li, the co-founder and CEO of SenseTime, mentioned that SenseTime will be committed to a balanced ethical governance standard of safe and controllable technology, people-oriented, and sustainable development, and advocated that the industry should uphold the “development” of artificial intelligence ethics.

The future direction of urban digitalization is very clear, that is, to continuously improve the quality of management and the breadth of management. In this process, the industry needs professional players such as SenseTime who have established a solid infrastructure in urban digitization, because the AI ​​technology ideas of such players have never left the specific real scene, and the solutions are also suitable Scenario needs, and the ethical standards of technology will be considered, which is the key to urban digitalization can be defined as high quality.

The Links:   LTM10C273 PT76S16