The rise of AI application is boosting the innovation of AI technology. Recently, the Chinese Academy of information and communications officially released the white paper on the development of AI framework (2022) (hereinafter referred to as the white paper), which defines the concept of rapidly emerging AI framework technology, combs the current situation of global AI framework development, and analyzes the future development trend of AI framework technology. The white paper points out that AI helps the current economy and society enter the era of intelligent economy. The world is entering a reshaping period driven by a new generation of information technology. As an important enabling technology, artificial intelligence has a strong "head goose effect" on activating the real economy and plays an important role in building the national scientific and technological influence. As artificial intelligence has become a new scientific and technological hotspot all over the world, the construction of artificial intelligence infrastructure has also become an important starting point and focus. The next decade will be a golden period for the global development of digital economy and intelligent economy society. Focusing on the development of artificial intelligence infrastructure will provide a strong traction for the development and growth of China's artificial intelligence industry and the vigorous development of digital economy. As a basic tool in the development of AI, AI framework plays the role of operating system in AI technology ecology. It is an important carrier of AI academic innovation and industrial commercialization. It will help AI from theory to practice and quickly enter the era of scenario application. It is also one of the necessary infrastructure for the development of AI. With the increasing importance, AI framework has become one of the focuses of AI industry innovation, which has attracted great attention from academia and industry. By definition, AI framework is a set of standard interfaces, feature libraries and toolkits for the design, training and verification of AI algorithm model. It integrates the encapsulation of algorithm, the call of data and the use of computing resources. At the same time, it provides developers with development interface and efficient execution platform. It is a necessary tool for AI algorithm development at this stage. AI framework can provide developers with the mathematical operation of constructing neural network model, convert complex mathematical expression into computer recognizable calculation diagram, automatically train the neural network and obtain a neural network model, which is used to solve the problems of classification and regression in machine learning and realize application scenarios such as target classification and speech recognition. According to the white paper, AI framework technology is just emerging, and the overall ecology of AI framework is far from mature in terms of technology evolution, open source ecology, market pattern, application empowerment, promotion and radiation. Hardware and software collaboration, open source building, developer promotion, empowerment in key areas and other aspects will provide important help for the ecological maturity and upgrading of AI framework. In terms of software hardware cooperation, AI framework technology should evolve from hardware adaptation to operator interface standardization. In order to cope with the challenges of diversification, complexity and fragmentation faced by the software and hardware ecology of artificial intelligence, the industry urgently needs to promote the hardware adaptation of AI framework and the standardization of operator interface. Encourage AI head enterprises to gradually build standardized hardware interfaces through the adaptation of AI framework and underlying AI chips, drive hardware manufacturers to actively adapt AI framework, and change from AI chip LED adaptation to unified hardware interface led adaptation. Support the development of unified AI operator interface standards, formulate standardized development interfaces by shielding different underlying hardware architecture details, and provide unified specifications for AI technology research, software and hardware development and application development. In terms of open source building, the industry should accelerate the implementation of infrastructure layout integrated into AI. The white paper believes that AI infrastructure is an infrastructure that can provide public inclusive and intelligent services for a long time with "data resources, algorithm framework and computing resources" as the core competence elements and "open platform" as the main enabling carrier. Encourage AI framework entities to provide AI capability services by integrating AI computing centers, AI public clouds and AI application open platforms, such as AI computing centers led by large and medium-sized cities, which is conducive to consolidating the underlying AI development capability. At the same time, it is necessary to support the large-scale model of deep empowerment and scientific calculation. Support the main body of AI framework to study the basic research field of scientific computing, and integrate super large-scale AI into the field of scientific computing through deep empowerment, so as to realize the rapid development of AI framework. Super large scale AI is the core driving force for the continuous transformation of AI in the past two years, and AI fusion computing is the key pillar for AI to enter various disciplines. Both of them are new highlands and new scientific devices in the field of basic scientific research in various countries. Super large scale AI and AI fusion computing put forward higher requirements for AI framework, and promoted the continuous improvement and upgrading of AI framework in technology from the multi-dimensional aspects of performance, accuracy, timeliness and energy consumption. At the same time, AI framework needs to actively integrate into the field of super large-scale AI and AI fusion computing, and carry out AI innovative applications, so as to provide rich resources for ecosystem suite and model base. (Xinhua News Agency)