Zhi DongXi (public number: zhidxcom)
Author | Yang Chang
Edit | Three Norths
Zhidong reported on April 1 that at the Baidu AI Open Day held yesterday, Ma Yanjun, general manager of Baidu's AI technology ecosystem, shared the three key points for the domestic deep learning framework to break through, as well as the development of Baidu's deep learning framework flying propeller.
Ma Yanjun said that according to IDC data, as of December 2021, Feipao ranked first in the comprehensive market share ranking of Deep Learning platforms in China, surpassing the deep learning framework launched by Google and Meta. Flying propeller attracts 4.06 million developers and serves 157,000 enterprises and institutions. At the same time, the flying propeller is also the first in terms of domestic chip adaptation.
First, the deep learning framework should break through three key points
As the backbone of promoting the intelligent landing of thousands of industries, the importance of deep learning frameworks is becoming increasingly apparent. The deep learning framework was included in the national "14th Five-Year Plan" and was included in the frontier areas of the new generation of artificial intelligence technology together with AI chips.
Ma Yanjun said: "In the AI technology system, the deep learning framework is in the waist position of connecting the upper and lower levels, connecting the chip and inheriting the application. ”
He mentioned that the current deep learning development is facing a double challenge, on the one hand, there are many application scenarios of deep learning, such as smart agriculture, smart manufacturing, smart education, etc., which require us to consider how to achieve large-scale application landing; on the other hand, deep learning needs to match the types of chips, such as NVIDIA, Intel, Baidu Kunlun XPU, etc., we need to consider how to make the deep learning framework run smoothly on various chips.
Since 2013, the AI academic community and industry have successively opened up their self-developed deep learning frameworks by institutions or enterprises, and built an AI open platform with the framework as the core, thereby promoting the formation of related industrial ecosystems. Among the more representative deep learning frameworks are Google's TensorFlow and Meta's PyTorch.
Ma Yanjun said that China's self-developed deep learning framework is gradually breaking through from international competition. However, China's self-developed deep learning framework still has a long way to go to gain a leading position in international competition. At present, the development of China's deep learning framework still needs to break through three key points: technical strength, functional experience and ecological scale.
In terms of technological innovation, the research and development of deep learning frameworks is inseparable from the underlying technical talents of AI, and the relevant talent reserves in China are not enough.
In terms of application experience, as the country promotes the digital transformation of the industry, the demand for AI applications in various industries is increasing. But sometimes, it takes at least 3-6 months for companies to apply a new technology, from the laboratory to the industrial landing. Therefore, a low-threshold or even zero-threshold development platform is particularly important.
In terms of application ecology, deep learning is a typical co-creation technology field, and only by building its own ecology can we achieve better sustainable development. However, the ecological construction cycle of the deep learning framework is long and the cost is high, and only when the framework is sufficient to meet the needs of developers, it is possible to attract developers and thus cultivate their own AI development application ecosystem.
Second, the flying propeller has gathered 4.06 million developers and adapted to 31 kinds of chips
Ma Yanjun said that Baidu Feipao has grasped the three key points to be broken through the deep learning framework through the practice of continuous technological innovation, putting the needs of developers first, and widely sharing with ecological co-creation.
Chinese industries and enterprises have their own characteristics, and the demand for AI is also unique. He added that if the domestic deep learning framework can not only meet the needs of Chinese industries and enterprises in terms of function, but also be simple and easy to develop, with a low threshold, then there will be a great opportunity to achieve curve overtaking on the industrial level, such as flying paddles.
Flying propellers have made great progress in the application of landing. At present, Feipao has gathered 4.06 million developers, developed 476,000 models, and accumulated 157,000 enterprises. According to IDC's deep learning framework platform market share report data in the first half of 2021, in China's deep learning market share ranking, feipao rose to the first place.
At the same time, Feipao has completed the adaptation and optimization of 31 kinds of chips with 22 domestic and foreign hardware manufacturers such as Baidu Kunlun Core, Huawei Ascend, Intel, NVIDIA, etc., covering the mainstream chips on the market.
Third, the AI base of Baidu's three major growth engines
At present, flying propellers have been applied in many industries, including smart cities, intelligent manufacturing, scientific research and other fields.
The application scenarios of flying propellers are also deepening, and the proportion of general scenarios such as face recognition in flying propeller application scenarios is gradually decreasing, while the proportion of key scenes in the industry is getting higher and higher, and the number of scenes in the forefront of the industry is also increasing.
Ma Yanjun mentioned that flying propellers are also the AI bases of Baidu's three major growth engines. Baidu's three growth engines are mobile ecology, intelligent cloud, intelligent driving and other growth plans. Taking Baidu Intelligent Cloud as an example, Flying Paddle can be called the core of Baidu Intelligent Cloud, bringing the blessing of AI technology to the intelligent cloud.
Conclusion: AI is empowering thousands of industries
From just open source to now, Baidu Feipao's cohesive developers have increased from 1.9 million in May 2020 to 4.06 million this year; the types of industries enabled are also more and more diverse, covering industrial, agricultural, medical, urban management and other fields.
On the one hand, the flying propeller goes deep into various practical application scenarios, making it easier for developers to use the flying propeller, on the other hand, it also actively adapts and integrates with various chips, forming the advantage of soft and hard collaboration. Based on these advantages, flying propellers have gradually become the first market share in the domestic deep learning market.
As the country promotes the digital transformation of the industry and all walks of life try to transform intelligently, the demand for intelligent technologies such as AI is increasing. We can see that AI is empowering thousands of industries.