Home >Technology peripherals >AI >For the first time, AI models have national standards: compiled by Huawei, Baidu, Peking University, Pengcheng Laboratory, etc., radiating AMD
News on April 11, on March 17, 2023, the national standard "Information Technology Neural Network" jointly compiled by 16 units including Huawei, Baidu, Peking University, and Pengcheng Laboratory Network Representation and Model Compression Part 1: Convolutional Neural Networks" (GB/T 42382.1-2023) was officially released, becoming the first national standard in China for artificial intelligence (AI) model representation and compression technology. It will be released on October 1, 2023 officially implemented on the day.
##According to reports, the national standard was developed by the cloud brain project team led by Professor Tian Yonghong, deputy director of the Network Intelligence Research Department of Pengcheng Laboratory, and belongs to the project team’s expertise in AI The systematic layout of global standards in the model field is an important achievement in promoting domestic standards in China. In December 2017, the Cloud Brain project team and core industry stakeholders officially launched standard development after the centralized research and development of key technologies such as AI model representation, compression, and deployment were mature. At the same time, in terms of standard promotion strategy, the Yunnao project team adopts the international and domestic "two-legged" model to iteratively optimize the standard based on feedback from relevant parties in different standardization organizations during standard development. IT Home summarizes it as follows:On the international front, the Yunnao project team took the lead in successfully publishing the "Standard for Artificial Intelligence (AI) Model Representation, Compression, Distribution and Management" at the Institute of Electrical and Electronics Engineers IEEE last year. , Compression, Distribution and Management Standard"), which based on the preliminary work, further combined with a wider range of opinions in the global field, formed the world's first international standard for AI models that is more adaptable to the international market, and thus won the Two IEEE SA standards awards. With the gradual deepening of research, the work of the Cloud Brain project team has attracted great attention from the industry, and it took the lead in successfully establishing a new generation of AI model representation technical standards at the Telecommunications Standards Bureau of the International Telecommunications Union ITU-T last year. Domestically, the Cloud Brain project team led the release of "Information Technology Neural Network Representation and Model Compression Part 1: Convolutional Neural Network" on October 21, 2021 at the New Generation Artificial Intelligence Industry Technology Innovation Strategic Alliance AITISA Network" group standard, and iterative optimization was carried out on the basis of the group standard, and finally the national standard released this time was formed. At this point, the layout of the first global AI model representation and compression technology standard system has basically taken shape.Based on the current needs of the AI industry in terms of resource-constrained devices and AI model deployment, this series of standards defines the representation specifications of AI models and the distribution, compression and deployment process to meet the needs of multiple scenarios. Breaking down the barriers between computing architecture and algorithm frameworks improves the efficiency of AI models in network distribution, learning, and application deployment. At the same time, some standards in this series have been widely used in fields such as AI development framework (Baidu), AI chips (Huawei HiSilicon), and AI open platforms (Hikvision), and the released series of standards radiate Go to AMD Xilinx and other international AI leading companies.
The above is the detailed content of For the first time, AI models have national standards: compiled by Huawei, Baidu, Peking University, Pengcheng Laboratory, etc., radiating AMD. For more information, please follow other related articles on the PHP Chinese website!