Home > News content

The quality of one hundred thousand industries is at that time: Huawei's cloud solved an industrial AI formula

via:博客园     time:2019/9/24 10:31:43     readed:144

How many industries are there in China?

According to statistics, China's industries are divided into 15 categories, 921 major industries and more than 100,000 sub-sectors. How many of these industries can be applied to AI technology?

By now, the answer has basically become a consensus: every one.

in fact, in fact, from

Last week, the Huawei full-link conference was a screen-brushing-level presence. However, the content is really too much, inevitably a bit dazzling. For example, you may not notice that the Huawei Cloud, which is closely related to the AI landing industry, has answered the industry AI through a series of industry readings and blockbuster releases in the Huawei full-link conference 2019 three days.

Taken together, we will find a formula for industrial AI. Huawei Cloud's various AI actions are based on the answer of this formula, let industry services evolve, let AI Development upgrade in an all-round way.

This is the key formula for the intelligent age, and that's why.

Industry AI

From the ability of logic and theory, to the integration and mutual promotion in the practical application, the story of AI and 100,000 industry has written only a few articles, even the preamble is not complete.

Mr. Zheng Ye, President of the Cloud BU of Huawei, used to discuss the issue with the General Assembly


President of Huawei Cloud Business Zheng Yelai.

Zheng Ye has proposed AI and industry integration for this purpose, and needs to solve the four major problems: to have a clear definition of a commercial scene, a strong computing force that can reach the hand and to provide the AI service of continuous evolution, and to meet the adaptation of relevant organizations and talents.

These four elements can be divided into two parts based on industrial logic. Half of them are upstream, that is, AI development; half of them are downstream, which is the combination of AI and industry. The upstream is more about the provision of technology and infrastructure, while the downstream is about the discovery and solution of specific problems in thousands of industries.

Back to Zheng Ye's four elements, we can change business and organizational changes to the industry's preparation for AI. It requires close cooperation between different roles and industrial ecology, and needs to be closely combined with industry wisdom and scene, the AI service of continuous evolution, and so on to solve the differential demand. Of course, with this Huawei cloud already in action, the establishment and development of the EI system is for this direction. In this all-link conference, the Huawei cloud formally released the EI industrial intelligent body, and also to solve the problem of the AI technology in-depth industry scene.

Here we press down the problem on the downstream side. Because the upstream is the basis of the AI fusion industry, the so-called foundation is unstable. When a lot of enterprises are facing the AI today, they haven't come to the step of in-depth scene innovation, just to understand the difficulty of AI from training to deployment, and immediately decide

Back to the four elements, what the upstream of the industry needs is the accessible calculation, the continuous evolution of AI service, and the ecological cultivation of talents.

This logic is actually very easy to understand, a company wants to use AI, so the most basic three conditions are: someone who knows AI, AI can quickly meet the standards to complete the work, the work has sufficient computational support.

So we can deduce the industry basis of industrial AI, depending on this formula: AI talent reserve

If we say that the downstream of the industrial AI is ten thousand problems solved by ten thousand people, the upstream is a problem solved by one person.

This is the industry observation and understanding of the Huawei cloud, and also the action guide of the Huawei cloud on the road of AI. At the full-join conference, we can find that Huawei's cloud is doing, that is to give talent, develop services, calculate the force, each link is put on one

Three-pronged approach is the correct solution to the problem of AI development.

Run together, burn together, have fertile soil, have developers

To promote the upgrading of the industrial AI efficiency, the first thing to do is to accommodate more AI developers to join the industry and to activate the change in the field of talent reserve.

What this matter needs is not just a tool, but an action. So we saw that at the Connection Conference, Huawei officially announced a 2.0-year development-oriented fertile soil project, announcing that it would invest $1.5 billion over the next five years. And a very important part of it is to attract enterprise and individual developers to join AI development ecology and become an ecological partner of Huawei Cloud to promote GSP AI.


Huawei Cloud

In response to the real needs of developers, according to the different stages of understanding, learning, building and listing between developers and AI technology, Fertile Land Plan 2.0 proposes to optimize the corresponding process, improve the quality of the community, and rationalize the action of resource allocation. To provide AI developers with lifting training cards, cloud service vouchers, OpenLab, training and certification services, developer competitions, etc. to build full-chain enabling from technology, ecology to business closed-loop.

From fertile soil project 2.0 to Huawei cloud enabling AI developers' specific actions, we can see a large number of industry-university-research integration, commercial closed-loop construction, developer empowerment implementation programs are being opened one after another. This full connection meeting, held Huawei Cloud garbage Classification AI Competition, Huafei Cloud Cup unmanned car competition final and so on, and released Huawei AI series first professional reference book "Shengteng AI processor architecture and programming in-depth understanding of CANN technology principles and applications." This series of actions is essentially to build an ecological community with AI developers to help developers understand Huawei technology and Huawei cloud AI. better.

The beginning of the Aesthetics of calculating Power:

Shengten910 opens the cycle of AI calculation

Man and arithmetic are the two ends of AI development. Talents have fertile soil, what is the calculating power? Huaweiyun's answer is violence.

On the first day of the full connection conference, Huawei released the Huawei cloud EI cluster based on Atlas 900, which has the largest AI computing power in the universe


Perhaps Huawei Cloud EI Clusters are not the capabilities common AI developers need to use. But its appearance really helped Huawei Cloud complete a high fight and took the industry high point of AI violent computing. Since the release of Shengteng 910 in August, we can see a series of capabilities and services based on Shengten 910 into Huawei cloud system. This change has occurred and will continue to evolve.

Based on the EI training cluster of Shengteng 910, Atlas800 AI server and Atlas300 training card joined Huawei cloud system one after another, which made a great breakthrough in the key computing power of AI training. AI training with high cost performance, high density, high concurrency and no long waiting for the trumpet is entering the developer's world.

AI development is a thousand-to-one thing, but AI training is only about two words, that is violence. After a series of hardware preparations have been completed, the AI calculated by the cloud of Huawei has entered the countdown.

The absolute soldiers of the gods wear automatic weapons:

ModelArts 2.0 Makes Development Qualitative

Why are AI development tired, cumbersome, inefficient, and most industries and businesses that sound interested in using it to crash? The reason here is neither lazy talent nor limited computing power.

A large number of redundant and repetitive training data tagging, long and low-precision black box training, chaotic model version management, reasoning deployment can not land a variety of compatibility problems, these are the real state of AI development, but also the most complex problem that AI is difficult to enter the industry. To some extent, AI development is a short-board project, any problem may stall the entire development process, resulting in applications difficult to land.

To solve this problem, we need AI services which are continuously upgraded iteratively, and improve the development efficiency with the evolution of infrastructure. At the Full Link Conference, Jia Yongli, General Manager of Huawei Cloud EI Services and Products, launched a one-stop AI development and management platform, Model Arts 2.0, to solve this problem.


Jia Yongli, General Manager of Huawei Cloud EI Service products Department

If the AI action of Huawei Cloud is to solve all the problems of AI development after disassembly, then Model Arts 2.0 is to solve all the problems in AI development technology process.

We know that AI development needs to go through four stages: data preparation, training, model management and reasoning deployment, while ModelArts 2.0 publishes more than 10 new features and services for all four stages, including intelligent data filtering, intelligent data tagging, intelligent data analysis, multivariate model automatic search, ModelArts SDK, graph neural network, reinforcement learning, model evaluation / diagnosis, model compression / transformation, automatic difficult to find. Online learning and so on, covering the whole life cycle of AI model.

Since the release of Model Arts last year, in fact, the positioning of this platform is to develop AI development toolbox with data automation and model management capabilities, so as to solve the problem that AI development depends entirely on programming ability and barbaric input. In this year's upgrade, the application of intelligent technology and a higher degree of integration automation constitute two upgrade curves, which intelligently deal with the waste of human resources in AI development. It will require more programming capabilities and higher threshold of work automation folding.


Such an upgrade takes into account the specific needs of different AI developers, and the Wulin expert can use the ModelArts 2.0 as a powerful tool to share their work and improve the development focus; new people can use the ModelArts 2.0 to directly develop and understand the AI development, which is a full-automatic weapon.

There are Shengong nearby, and automatic fire protection body, it can be called the standard of Chinese captain Yan Shuangying.

Let's take a few examples to see how the ModelArts 2.0 makes the development a qualitative change:

1. Intelligent support for data processing.

Data tagging and preprocessing is the most tedious process in AI development. Although some semi-automatic tools can improve the efficiency of developer tagging, if invalid data is mixed, it may reduce the training accuracy and play a negative effect. In order to solve this problem, ModelArts 2.0 upgrades the intelligent data filtering function, and automatically filters and filters out the data which are invalid to the training model by AI, such as defocusing, overexposure, not meeting the training requirements and so on. In addition, ModelArts 2.0 also upgraded hybrid intelligent tagging, intelligent data analysis and other capabilities. Having AI prepare you for AI data has a wonderful sense of harmony.

2. AI training, from traditional to modern.

In the process of AI training, how to improve the training accuracy has always been the core problem of testing developers. Through Huawei cloud's original multiple search technology, automatic data enhancement, super parameter automatic search, neural network architecture search and other technologies are integrated. ModelArts 2.0 can help users build AI model quickly, and the precision performance is much higher than that of the industry. This kind of highly intelligent minimalist training program can be said to make the training move from tradition to modern.

3. Model management and new package integration.

In the face of AI moving to industry, it may go through complex model management engineering and high precision and high security standard version management. ModelArts 2.0 also upgrades a wealth of evaluation interface and visualization ability, so that the AI model can be seen at a glance that it will no longer encounter confusion as soon as it arrives at management.

4. Reasoning deployment and the evolution of necessary capabilities at the industrial level.

In the field of AI deployment, continuous evolution and continuous learning of models is actually a necessary requirement for industrial applications. In the industrial AI scenario, the model must have the ability to continue to grow, iterate, and smooth the transition. In order to solve this problem, ModelArts 2.0 industry first difficult automatic discovery function, combined with online learning mode, open up the running state and development state, so that the model changes with the growth of data, which also means that the model can continue to learn and evolve online. This feature has been tested on Huawei cloud autopilot cloud service Octopus, which also shows the direction and goal of ModelArts becoming the core platform of industrial AI.

The in-depth automation and integration make AI's technical difficulties be folded, so that AI can really be used; the advanced intelligence makes AI technology do to reduce the work burden of experts, let human intelligence return to creation and innovation, and let AI development go deep. These two encounters make AI development truly qualitative.

Huawei Cloud AI: Why is it at the time?

Let's go back to the original formula and see that EI Training Cluster, Fertile Earth 2.0 and Model Arts 2.0 point to three key points for AI developers to upgrade. The combination of the three triggers a qualitative change in the cost of AI development time.

Empowerment of AI developers means introducing more talents and shortening project time through talent cooperation and organization.

The AI computing cloud service based on Shengten910 chip will bring more abundant AI computing power and realize shorter mechanical training time.

The toolbox of ModelArts 2.0 has been upgraded in an all-round way, which reduces the complexity of the whole AI process and shortens the manual development time.

The shortening of all time ultimately means that the feasibility of the industry will be enhanced, and the efficiency of AI development infrastructure will be completely reformed. Ai will no longer be a difficult choice to carry out. Industries and companies can focus on business closed-loop and innovation value, rather than losing their heads in the time maze of AI development.

Be friends of Pratt & Whitney AI and users. Huawei Cloud's real answer to this formula is that industrial AI is making a qualitative change in productivity, so it's time to choose Huawei Cloud.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments