Wang Haifeng, | of information flow: Artificial intelligence presents the characteristics of integrated innovation and lowering the threshold at this stage

On December 12, the WAVE SUMMIT+2021 Deep Learning Developer Summit hosted by the National Engineering Laboratory for Deep Learning Technology and Applications was held in Shanghai. Wang Haifeng, chief technology officer of Baidu and director of the National Engineering Laboratory for Deep Learning Technology and Applications, released the latest report card of Feipao: condensing 4.06 million developers, creating 476,000 models, serving 157,000 enterprises and institutions, ranking first in the comprehensive market share of deep learning platforms in China.

He said that the flying propeller industry-level deep learning open source open platform is a typical AI large-scale production platform, empowering the majority of developers, strongly supporting the large-scale production of AI industry, and promoting technological innovation and industrial intelligent upgrading.

Flying propeller is a typical AI large-scale production platform

At the first summit in April 2019, Wang Haifeng proposed that deep learning promotes artificial intelligence to enter the stage of industrial large-scale production, and the deep learning framework is connected to the chip and the application is inherited, which is equivalent to the "operating system in the intelligent era".

In the large-scale production stage, it is necessary to have an AI large-scale production platform, which can output AI technology to all walks of life in a standardized, automated and modular manner to achieve large-scale application; at the same time, it can unite all parties and integrate innovation and develop together on the basis of the platform.

In the past three years, facing the needs of technology and industrial development, Feipao has continued to break through the core framework of innovation on the one hand, realizing the convenient development, efficient training and multi-terminal multi-platform deployment of AI models; on the other hand, Feipao is open source and open, gathering developers and all sectors of society to build and strengthen the ecology, and cultivate AI talents to achieve mutual promotion of technology, industry, talent and ecology.

Characteristics of the AI large-scale production stage: "Convergence of innovation, lowering the threshold"

Based on the observation and thinking on the development of AI technology and applications, as well as the practice of flying paddle industry-level deep learning open source open platform, Wang Haifeng believes that at present, artificial intelligence presents the characteristics of "fusion innovation" and "lower threshold": on the one hand, there are more and more integrated innovations in AI technology and industry; on the other hand, although AI technology is becoming more and more complex, the threshold for AI development and application is getting lower and lower.

In terms of the integration of knowledge and deep learning, Baidu has just released the world's first 100 billion big model of knowledge enhancement - Pengcheng - Baidu Wenxin, with a parameter scale of 260 billion, achieving the best results in more than 60 tasks. The release of the 100 billion large model of knowledge enhancement also benefited from the large-scale training technology of flying propellers. Flying Propeller has developed an adaptive large-scale distributed training technology, which makes the training efficiency of large models reach the best level in the industry.

In terms of cross-modal multi-technology integration, machines need to understand the real world like humans, understand speech, understand images and videos, and understand language. Baidu has developed a large model of knowledge enhancement across modes to improve cross-modal understanding and generation capabilities.

In terms of software-hardware integration, the chip and deep learning framework are jointly optimized by combining factors such as computing power, power consumption, and latency to obtain the best results of AI applications. Flying Propeller has completed the adaptation and joint optimization of 31 chips with 22 domestic and foreign hardware manufacturers.

In terms of technology and scene integration, in the process of application to all walks of life, flying propellers are integrated with practical application scenarios, which not only effectively solves the industry application problems, but also enables the flying propeller platform to be continuously accumulated and improved. Feipao also launched a series of scientific computing APIs to support applications such as quantum computing and life sciences, and explore the integration and innovation of AI and cutting-edge technology.

Although AI technology is becoming more and more complex, the threshold for AI development and application is getting lower and lower.

In terms of technology, it has recently become a large model of important direction of AI technology, with the characteristics of good versatility, strong generalization, and good effect, which can reduce the threshold of AI development and application. The dynamic and static unity realized by the flying propeller on the core framework, taking into account the flexibility of scientific research and development and the efficiency of industrial development, can meet the needs of different types of developers.

In terms of tools and platforms, Feipao has created tools such as inference deployment tool chain, model compression, and industry-level model libraries to reduce model development and deployment costs.

In terms of ecology, Feipao has prepared an industrial practice case database for developers to help enterprises learn from experience and inspiration and accelerate intelligent upgrading; build a Feipao artificial intelligence industry empowerment center to accelerate the landing of artificial intelligence with partners.

Cultivate compound AI talents

With the deep integration of AI and industry, more and more composite AI talents who understand AI and have industrial experience are needed. All along, Feipao has been committed to the cultivation of composite AI talents, uniting academia and industry, opening up the positive cycle of production, education and research, and continuing to contribute AI talents to society.

At the summit, Feipao released ten latest developments, including the new panorama of Feipao, the co-creation plan of Feipao "Great Navigation" 2.0, and the industry's first industrial practice sample library.

As Wang Haifeng said, Feipao has always been adhering to the original intention of technological innovation and open source, keenly judging the development trend of technology and industry, working the accumulation and breakthrough of core technologies, and at the same time solidly and steadily promoting ecological construction, exploring and growing together with the majority of developers, industry, education and research parties, and empowering industrial transformation and upgrading.

Upstream journalist Yang Ye