Li Gen
reports from ShanghaiAI Quantum Bit | Public Account QbitAI

Huawei is a "scary" company.

Whatever they are doing and want to do, once they start to execute, will always be amazing, whether it is old business or new field.

For example, in the cloud business, the outside world’s attitude has changed drastically within a year.

When the BU was established and launched to the public last year, some were optimistic and fully supportive; others were pessimistic, thinking that the market structure for cloud business had been determined, with many Chinese and foreign giants competing against each other, and that it would be difficult for Huawei to achieve the same success as with smartphones.

But so far this year, the left hand has handed in the report card, and the right hand has launched a new big move.

After the 2018 Huawei Connect Conference, the momentum of Huawei Cloud is worth re-examining.

Huawei Developer Conference

Let’s start with this year’s HC conference.

HUAWEI CONNECT, often referred to as Huawei HC Conference , is an annual conference for Huawei customers and partners.

However, this year, the trend has changed dramatically. Whether it is the pre-sale of tickets or the on-site participants, the number of technical practitioners and developers is unprecedented.

So there was a joke at the scene: the first "Huawei Developer Conference".

The result is indeed appropriate. At this year's conference, Huawei not only announced its AI strategy and released AI chips, but also launched a set of solutions defined by Full stack and full scenario , which are aimed directly at AI developers.

Huawei itself also said that it does not admit that this is a "transformation" because what it has done is just "moving forward" towards new technological trends.

According to Huawei's rotating chairman Xu Zhijun, AI is a general technology that will inevitably affect all aspects of the future. Huawei's solutions and strategies, whether for internal improvement or external customer needs, are implied - and even the only choice.

Only after this decision did Huawei Cloud's status and mission become clearer.

A year ago, Huawei Cloud was just starting out, and Huawei Cloud was more like a bridgehead. But after the AI ​​strategy, the Black Soil mission was put forward at the beginning, and the path could not be clearer.

Moreover, Huawei Cloud will not only be the fertile soil for its customer partners, but also serve as an incubator for AI developers.

You say, isn’t the importance more self-evident?

The playing style is also becoming clearer.

Huawei Cloud Advanced

In fact, there were signs of Huawei Cloud's advancement when the BU was established last year.

It is verifiable that last year, Huawei Cloud's EI business, which took up the banner of AI pioneer, only had "tasks" but no KPIs.

It is said that the top management only requires: to go deep into the industry, to have more contact with customers and partners, to succeed in every project and to do it well. But there are no specific evaluation indicators.

However, judging from the transcript, it can be called excellent. After all, entering eight industries and completing more than 200 projects in one year, many of which are high-profile and large-scale cases, is not easy.

For example, projects such as Shenzhen Traffic Police and Beijing Xi'erqi in the transportation field, and KingMed Medical in the medical field.

But Huawei Cloud itself reviewed and believed that the more important gain was that it had extensive exposure to various industries, laying the foundation for the launch of Huawei EI City Intelligence .

What is urban intelligence?

That is, on the basis of digital twins, artificial intelligence is used to coordinate cloud, big data, edge computing, and Internet of Things technologies to form a complete system from data generation, analysis, and closed loop.

Ultimately, we will leverage the powerful model building capabilities and computing power of the digital world to drive the physical world towards intelligence.

To put it more bluntly, it is to treat the real space that was difficult to "calculate" in the past, especially cities involving various businesses and dimensions, as a whole, and then be able to model and calculate it, and use AI and other methods to achieve the connection between the virtual world and the real world, and allow the real world to iterate day and night like the virtual space and become intelligent.

Similar visions and ideas are not unfamiliar.

However, it is easier said than done. To truly put it into practice, we need to first succeed in a series of vertical solutions such as security, transportation, water conservancy, and water affairs, and find a model solution for rationally allocating resources and efficiently scheduling resources.

Therefore, the launch of the urban intelligent body at this time also means that the above-mentioned vertical scenarios, as well as the previously launched Huawei Cloud EI traffic intelligent body, industrial intelligent body, etc., have tasted the sweetness.

In addition, Huawei’s AI strategy has been clearly defined, and the idea of ​​“platform + AI + ecology” is clear. Moreover, from now on, Huawei’s biggest difference is: + chip .

Huawei Ascend, as the main AI product, can not only bring about an improvement in computing power, but also, due to its independent design and definition, it can better take into account the needs of business scenarios.

For example, it took 18 minutes to complete a general ImageNet dataset training using 128 GPUs, but with Huawei Cloud EI, the same result can be shortened to 12 minutes.

On the other hand, there are supporting architectures and system-level services. So after the release of the AI ​​chip, Huawei Cloud immediately announced a service upgrade: Huawei Cloud EI will provide platform basic services, general APIs, high-level APIs, and pre-integrated solution services, with a total of 45 services and 142 types of functions online.

This package of services claims to cover developers including data scientists, data algorithm engineers, IT developers, and even business personnel without AI background.

Therefore, Huawei Cloud itself also said that after adding chips and upgrading, it has entered a new era of AI.

AI Development Tools

But is this the entire ambition of Huawei Cloud?

Not really.

The AI ​​full-process development platform and tools released this time may have a wider impact :

• AI development platform ModelArts

• Visual AI application development platform HiLens

• Quantum computing simulator and programming framework HiQ

This combination of punches, every move directly hits the pain points of AI developers, especially the AI ​​development platform ModelArts .

ModelArts covers the entire AI development process, including data labeling and preparation, model training, model tuning, and model deployment.

For example, in the data labeling and preparation stage, which currently requires the most "manual" work, ModelArts has a built-in AI data framework, which uses AI mechanisms to manage data and uses iterative training to solve the problem of the amount of labeled data. It is especially suitable for scenarios with large amounts of data, and the efficiency of data labeling and preparation will be increased a hundredfold.

In the model training phase, the most troublesome issue is the training time. ModelArts can reduce the model training time by half under the same model, data set and hardware resources by improving computing power and computing solutions.

Perhaps it is because Huawei is a large company with a strong background in technical research and development that it is so familiar with the pain points of AI development.

Zheng Yelai, President of Huawei Cloud BU, also said that the main contradiction in current AI development is the contradiction between AI developers' long-term demand for universal computing power and the current scarce and high cost of computing power.

This is true.

Therefore, the AI ​​development platform launched by Huawei Cloud aims not only to help developers get started faster, train faster, and deploy faster, but also to take into account inclusive AI - affordable, easy to use, and safe to use.

The same inclusive AI concept is also reflected in the visual AI application development platform HiLens .

HiLens consists of a camera with AI capabilities and a cloud development platform. SKILL consists of control code and models. The AI ​​model is trained by ModelArts. HiLens is also compatible with models trained by other mainstream frameworks. The SKILL developed and generated by HiLens can be directly deployed to any end-side device with Ascend chips.

For example, the camera equipped with Ascend 310 can process 100 frames per second and realize millisecond-level face detection. At the same time, it also has a built-in lightweight container, which can achieve less resource usage, reduce network bandwidth, and fast download startup time.

△ Camera equipped with Ascend 310

HiLens can play a role in scenarios such as robots, fatigue driving detection, production safety detection, and logistics. It is like equipping AI applications with smart eyes, so Huawei also gave HiLens the code name: Smart Eyes.

At the conference, Huawei Cloud also attracted attention with a "robot dance".

It’s just that it’s not common for robots to learn to dance like humans, and the whole “learning” process is very clear:

1) Use ModelArts to train a "human key point detection" model;

2) Import the trained model into the HiLens platform;

3) Select the corresponding video to learn and deploy it to the robot terminal;

4) The robot begins to perform the human dance learned from the video.

Third, there is the future-oriented quantum computing simulator and programming framework HiQ .

In terms of design, HiQ not only has full-amplitude and single-amplitude simulators, but also takes the lead in the industry to launch: high-performance quantum error correction circuit simulator cloud service - which can simulate hundreds of thousands of quantum bit circuits, in addition to the classical-quantum hybrid block programming user interface BlockUI.

Among them, HiQ can simulate more than 42 qubits with full amplitude and more than 81 qubits with single amplitude. For low-depth circuits, single amplitude can simulate 169 qubits. This is the industry-leading quantum circuit simulation cloud service.

Finally, Huawei Cloud also offers a sugar-coated bullet to AI developers by launching the fertile AI developer enablement plan .

Let’s take a look at the key points:

• For developers: provide 20 hours of free training, 21-day AI training camp, host AI developer competitions, and provide exclusive innovation incubation camps.

• For partners: Provide 1,000 sets of free development environments, including development modules and boards; provide expert resources for the first batch of 20+ key partners.

• For universities and research institutions: Invest RMB 1 billion in AI talent training, including providing Huawei Cloud AI resources and AI suite support; also collaborate with universities and research institutions to develop AI courses.

At present, the Institute of Cross-Disciplinary Information Sciences of Tsinghua University, University of Science and Technology of China, Zhejiang University, Shanghai Jiao Tong University, Nanjing University, Southeast University, Xidian University, Chinese Academy of Sciences and others have announced on the spot that they have joined Huawei's AI talent training program.

Re-examining Huawei Cloud

With all the above plans and policies, I wonder if you will be impressed by the prospects of Huawei Cloud.

But at the HC conference, participating developers expressed their opinions with participation and applause.

At the exhibition site, Huawei Cloud has specially opened a "practical area". In addition to experiential interactive areas for AI solutions in transportation, retail, etc., there are also laboratories where developers can actually get started and experience low-threshold AI development on site.

In fact, it is also lamentable that China already has the most enthusiastic and potential AI developers in the world, but has always lacked a development environment and support to match them.

Now, Huawei Cloud has stepped forward, with chips and architecture, as well as a series of AI development platforms and tools, demonstrating its sincerity and starting to take action.

So I hope Huawei will continue with its usual style.

If successful, it will not only be a blessing for Huawei, but also for AI developers.

How should I put it? I hope you will not forget your original intention and keep moving forward (serious face)~