Electronics

"Only when the application is implemented, the underlying AI chip and software technology can be more solid in the process." Ma Yanjun, senior director of Baidu's deep learning technology platform, said in a speech at the first Global Digital Economy Conference-Smart Core Development Forum held recently. , It takes at least 3-6 months to complete a project with artificial intelligence end-to-end. The process is complicated and requires a series of guides for selection schemes. As the first batch of member units, Baidu, Cambrian, Suiyuan Technology, Cloud Control Bee Core, and Ling Yunguang jointly participated in the forum and conducted AI- Rank strategic cooperation signing ceremony. AI-Rank emphasizes application-driven, and aims to truly and objectively reflect the situation of the artificial intelligence industry through the establishment of an open source benchmark test system.

AI small image.png

Artificial intelligence is a technological "highland" that countries all over the world are vying for. In order to avoid the barbaric growth of the industry, management departments, industry organizations, and associations of various countries have continuously increased their standardization work to promote the healthy and sustainable development of the industry. Although my country's artificial intelligence industry has developed relatively late, it continues to keep up with the global frontiers in technological exploration and marketization, and strives to be in line with international standards in standard formulation.

AI commercial landing calls for standard packages

"Only when the application is implemented, the underlying AI chip and software technology can be more solid in the process." Ma Yanjun, senior director of Baidu's deep learning technology platform, said in a speech at the first Global Digital Economy Conference-Smart Core Development Forum held recently. , It takes at least 3-6 months to complete a project with artificial intelligence end-to-end. The process is complicated and requires a series of guides for selection schemes. In the past, the industry focused more on algorithms. In recent years, algorithms have begun to go hand in hand with engineering open source, and more and more attention has been paid to the collaborative development of software and hardware. Artificial intelligence, big data, cloud computing and other new information technologies have become more and more integrated.

The "Artificial Intelligence Standardization White Paper (2021 Edition)" released in July this year pointed out that the deep learning framework's reliance on ecological construction and the incomplete testing system are two major problems currently encountered in the development of my country's artificial intelligence industry. my country's deep learning framework started late, and it has not yet got rid of its dependence on foreign deep learning frameworks in terms of algorithms, chips, terminals and scene applications. Baidu has seen such development pain points and positioned its artificial intelligence development strategy on the exploration of deep learning frameworks. It will integrate its open source deep learning platform flying paddle model with business partners such as COMAC, Industrial and Commercial Bank of China, CATL, State Grid, etc. Adapt to promote the application of local open source frameworks. For example, CATL used flying propellers to upgrade the quality inspection production line of new energy batteries. It is understood that compared with traditional detection algorithms, the pass-kill rate is reduced by 66.7%, the algorithm generalization ability, deployment efficiency is comprehensively improved, and product costs are greatly reduced.

However, the domestic artificial intelligence test system has not yet been formed, the test content and models of the existing test benchmarks are highly repetitive, and mature functional and performance test benchmarks have not yet formed, which will restrict the artificial intelligence products from opening the market and gaining market trust.

Industry supervision urgently needs to improve AI standards

"Standards, especially those established by existing international standards bodies, can support the global governance of artificial intelligence development." Peter Cihon, an assistant researcher at the Artificial Intelligence Governance Research Center of the Institute for the Future of Humanity, Oxford University The technical report of "International Standards for Global Cooperation in Artificial Intelligence Research and Development" pointed out that standards will not achieve all artificial intelligence policy goals, but they are the path to effective global solutions.

In recent years, the process of artificial intelligence standardization in various countries around the world has been continuously promoted. At the policy level, attention to industry standards and governance has increased year by year, and the global artificial intelligence standardization work has entered the stage of accelerated implementation. In January 2021, the "American Standardization Strategy 2020" issued by the American National Standards Institute (ANSI) shows that the United States is currently facing more and more challenges in setting new standards, as well as horizontal issues such as artificial intelligence privacy and security. These key issues Standard formulation requires vigorous cross-departmental cooperation; in April, the European Commission’s Joint Research Center released the "Artificial Intelligence Standardization Pattern—The Relationship between Progress and the Proposals for an Artificial Intelligence Regulatory Framework" to support artificial intelligence supervision through the formulation of international and European standards.

Peter Cihon mentioned in the report that standards can influence the development and deployment of specific artificial intelligence systems through product specifications, that is, explanatory, robust and fail-safe design. They can also affect the research and development of artificial intelligence and the general environment for deployment through process specifications.

Accelerated domestic AI standardization process

Standardization is one of the keywords for the current development of domestic artificial intelligence. Statistics from CCID Consulting show that in 2019, the scale of China's artificial intelligence industry reached 129.14 billion yuan, a year-on-year growth rate of 30.8%. It is estimated that by 2022, the scale of China's artificial intelligence industry will reach 262.15 billion yuan. "No rules, no squares", the domestic artificial intelligence industry is developing rapidly, and scene applications are gradually enriched. As the complexity of the industry increases, related standards also urgently need to solve supporting problems.

In order to strengthen the top-level design of standardization in the field of artificial intelligence, promote the research and development of artificial intelligence industry technology and standard formulation, and promote the healthy and sustainable development of the industry, the National Standardization Management Committee, the Central Cyberspace Administration of China, the National Development and Reform Commission, the Ministry of Science and Technology, and the Ministry of Industry and Information Technology Ministries and commissions jointly issued the "Guide for the Construction of National New Generation Artificial Intelligence Standard System" in July 2020. In December of the same year, Zhongguancun Intelligent Artificial Intelligence Research Institute (hereinafter referred to as "Intelligent Research Institute"), Baidu and Inspur jointly issued AI-Rank, an open source evaluation benchmark of artificial intelligence for industrial applications. The evaluation system adopted by the benchmark can pass multiple index tests to comprehensively score, aiming to objectively, comprehensively and systematically reflect the current situation and development direction of the artificial intelligence industry. At the first Global Digital Economy Conference-Smart Core Development Forum held in Beijing a few days ago, five companies, Baidu, Cambrian, Suiyuan Technology, Cloud Control Bee Core, and Lingyunguang, were the first batch of member units, and intelligent artificial intelligence. The institute carried out the signing ceremony of AI-Rank strategic cooperation. Sun Mingjun, executive director of the Zhongguancun Intelligent Artificial Intelligence Research Institute, said that in order to be able to truly and objectively reflect the industry situation, AI-Rank emphasizes that it will be driven by applications and will meet the requirements of openness, fairness and fairness through the establishment of an open source benchmark test system. The test results can be quantitative, reproducible, and comparable, and attract community developers to build together. "I hope to establish a standard and certification system with international leadership and credibility to promote the industry to take off." Sun Mingjun said.