Huo Houkun, the chairman of Huawei rotation: Promote the fusion of artificial intelligence advance and computing power network, accelerate the incubation and innovation of industry applications

21st Century Business Herald reporter Xu Qiulian, intern Pan Jia Chuan Shanghai report

On September 1st, the Fifth World Artificial Intelligence Conference (WAIC) officially opened. At the opening ceremony, Huawei Chairman Hu Houkun said that the key to the development of the next generation of artificial intelligence is to build a good strategic facility for artificial intelligence , Plan the direction of application innovation, and truly allow artificial intelligence to create value for the digital upgrades of all walks of life.

The demand for artificial intelligence computing power in the next 10 years will increase by more than 500 times

When talking about computing power, not only the network computing power is very important, but the network is also very important. Hu Houkun predicts that the future computing power network is like today's power grid, communication network, and high -speed rail. It can provide strong impetus for the national digital economy and the development of the entire economy.

Regarding the construction and deployment of the computing power network, he suggested that the actual situation of China's actual situation is based on the actual situation of China. According to the prediction of the computing power demand of artificial intelligence in the next 10 years, it will increase by more than 500 times, becoming the largest increase in the future computing power. It can be completed from this increment. The network of Li, this is also the first test for the country's implementation of the eastern and western strategy.

Secondly, computing power network is moving towards fusion heterogeneous. Not only the network of artificial intelligence computing centers, but also the supercomputing centers and integrated big data centers in various places can incorporate this computing power network to form a large -scale large computing power platform to support the high quality of the digital economy to support the digital economy develop. He believes that it should vigorously promote innovative national network technology, and build a network that can provide larger bandwidth, lower latency, and highly certain connection network to ensure efficient scheduling of data applications and algorithms.

He suggested that all parties should work together to build a relatively unified standard. Specifically, at multiple levels of hardware application interface, node interconnection and data sharing, the standards of the standards are achieved. "In this way, the computing power data and ecology can be converged, and the sharing and efficient operations of the entire network can be achieved.

It is not enough to have a unified standard. Hu Houkun also emphasizes that the basic software ecology of the development of artificial intelligence is also a key support for operation. While vigorously developing hardware such as chips and networks, we should pay attention to basic software, such as artificial intelligence framework, development kit, and development kits. The coordinated development of basic models and the computing power of the release of hardware, eventually allowing artificial intelligence to land in more industries.

Direct the incubation and innovation of different industries with "big models"

In addition, under the premise of improving network computing power hardware and software architecture, Hu Houkun believes that it should further accelerate the incubation and innovation of industry applications and make artificial intelligence technology play greater value. “当前孵化大模型已经成为行业与场景创新突破的一种共识，如从药物研发来说，场景是复杂多样的，比如说蛋白质小分子结合的预测，小分子属性的预测以及小分子的优化和For various scenarios such as generating, if each scene goes alone to build artificial intelligence training models, its efficiency is very low and the cost is very high. And through a unified "Pangu Pre -training" model with a unified pharmaceutical industry, it can be based on oversized large models. The scale of the scale, the data of massive training, and the needs of the drug development of multiple different scenarios, thereby greatly shortening the cycle of drug research and development. "

However, the research and development threshold of large models is high and time -consuming and labor -intensive. Pay attention to avoid repeated investment and development. In this regard, he called on politics, production, learning, research, use, etc. to combine the basic big models and industry models required by different industry scenarios together, planning large models of sand tables, and incubation and innovation of large models. It can reduce repeated investment, but also conducive to concentrating advantageous resources, and jointly accelerate the penetration of artificial intelligence applications to various industries and industries.

In addition to the development of the algorithm of the big model, Hu Houkun also emphasized that it is necessary to combine the knowledge of various industries in order to make artificial intelligence landing the application of various industries. By establishing an industrial consortium and gathering the forces on the industrial chain, it will vigorously improve the implementation of artificial intelligence applications. efficiency. For example, the two industries of remote sensing and multi -mode state have been built and blossomed, which attracted more than 70 scientific research institutions and enterprises to join, hatched new applications of more than 20 industries, and promoted large artificial intelligence models. Remote sensing, simulation, finance and other industries.

- END -