A few days ago, Huawei and Xinzhiyuan hosted the Huawei HiAI Open Open Course. Many of our readers paid attention to this course. So many readers left a message or invited us to answer in the Q&A platform. How to treat and understand This course also has readers who are developers or practitioners of mobile applications, asking us how to combine the AI development capabilities brought by HiAI platform with our own business, and want to take this 'Road to Mobile AI'. What to pay attention to.
Think about it, this is indeed a detailed explanation of the "dry goods sharing" of HiAI architecture and Huawei's proposed mobile AI strategy in various aspects. Interested readers may wish to check it out. But after all this is an AI-oriented platform and technical ability. Sharing, Many mobile application developers may be relatively unfamiliar with this area, so we come today to answer some readers' questions.
Most of the mobile application developers I came in contact with are actually very good at hearing about AI. However, it is still somewhat ambiguous as to how to enter AI, how to quickly complete commercialization, and how to choose how to connect with the platform. .
So we hope that in the most straightforward way, from the point of view of a developer who does not understand the algorithm (as is the case with most mobile developers in China), we will examine what the HiAI architecture has brought developers in the end. At the same time, how should developers think about how to complete the fast AI of their own app?
From the mobile era to the AI era, you can't listen to what other people are saying, and just dance with the tune. More importantly, developers need to know what exactly 'I' really wants.
'I want a small risk to try'
If we are mobile application developers, we have heard that AI can implement various functions for its own APP and blessed with various cool games. What is our first reaction? In fact, the most rational response must be: Can you afford it?
This is not a matter of worrying people. Just now, AI God Yann Lecun made a tweet, which means that the AI bubble will break down quickly. The core reason is that many targets are set at an AI company that is as high as the moon. The money is running out soon....
Probably the greatest tragedy in the world is this: AI hasn't figured it out, money hasn't worked out yet...
In fact, before the arrival of the HiAI architecture and the Kirin 970 chip, if the Chinese mobile application developers want to try AI, the basic routine is to purchase cloud service providers' AI-related computing power and services. This is based on traffic accounting, and the more users the cost The higher, and the cost of many test deployments, is even more horrific. Developers must collect massive amounts of training data from scratch, as well as the enormous workload and training difficulty of model training.
Obviously a small developer, the entrepreneurial team is very hard to bear the time, technology and cost of developing AI applications from scratch. The medium and large mobile application teams cannot afford the uncertainty of value and strategic squeeze.
So, if I am a mobile developer and want to try AI, but I am not 100% sure about the future, what do I need most?
The answer is, the opportunity to quickly enter the field at a small cost.
The reason why the HiAI architecture is currently unique and ahead of the industry's imagination is because the open-end AI computing capabilities are only one in the industry. In other words, developers avoid the high cost of AI development using cloud services. Based on the special processing capabilities of the NPU in the Unicorn 970 chip, developers can enjoy 10 times higher AI computing power than the GPU, and can avoid high cost pressures.
After connecting the HiAI architecture, Huawei's mobile AI ecosystem provides developers with a relatively complete set of five major engines and a full set of interfaces. This gives developers a platform to achieve specific AI capabilities, avoiding their own collection. Data, the technical difficulty of training this process from scratch and a lot of time and money costs.
From the perspective of developers' value selection, the emergence of Kirin 970 and HiAI architecture is based on the relatively more rational calculation support of end-to-end computing, providing platform support for developers and solving hardware bottlenecks. A fully open platform The output of the strategy solution allows developers to save everything from scratch, focus on mobile scenarios, and focus on the AI experience on mobile phones.
Taken together, after the HiAI system with low risk, low threshold, and strong ecological integration is built, application developers can develop innovative AI applications based on platform capabilities instead of starting from scratch and facing the unknown. High cost long blind exploration.
Risk is small, return can be expected, is the premise of all technical business.
'My APP, of course I'm the master'
In the face of AI, another developer must pay attention to whether their own APP can achieve growth through AI, or just make it fun?
In many fields today, the experience of using AI scenarios or AI capabilities alone is already well-known. For example, the purchase of maps in the field of e-commerce. Once a certain ability is known, the general situation is that the competing products in the industry will follow suit to join in. Many developers actually have to 'accept AI'.
However, there is a problem here. This AI application ability of follow-up and imitation is actually only a single fragment. Everyone has, of course, they have to follow. However, based on the occlusion of the development environment, this AI capability cannot be improved. Can not be linked with other functions. Over time, developers will find themselves a lot of manpower and resources by an AI function, after the version is updated, nothing can be done, can only let AI become tasteless.
This is the problem that the development platform can't solve the problem of correlation reasoning and continuous development. Because the completed models in the machine learning framework are all single, it is difficult to join together with other capabilities to form a whole.
The solution to this problem is to use the reasoning and development capabilities of the HiAI architecture. HiAI architecture currently provides solutions and platform functions, covering video, camera, AR, e-commerce, social networking, and language translation. In large areas, it can be said that it basically covers the major areas of today's mobile applications, and has fully opened up chip capabilities, application capabilities, and cloud capabilities.
In other words, the development of a variety of related skills, or different degrees of development and upgrading, can be achieved with HiAI, to achieve the strong growth of the application.
This is a question worth paying attention: Many mobile developers who have rushed into the AI world are only satisfied with a certain detail AI. This actually gives the user a limited increase in the experience, but it costs a lot of their own costs. AI's The premise must ensure that you are the owner of the APP. You can clearly plan the needs and development of the next step of the APP. There is no gap between the idea and the implementation.
In the development of application layer capabilities that are highly emphasized in the HiAI architecture, a general-purpose deep learning development framework is integrated, which is compatible with various development methods. This means that developers will not only complete acceleration on HiAI, or achieve AI of a certain capability. Then there is nothing to do. HiAI provides a capability integration service that can combine recognition, learning and proactive output of multiple capabilities, allowing developers to find the right aggregation point in the AI field.
A good platform, of course, does not give developers a way to get the developer to go dark, but should give the developer a board, letting everyone go in and out.
'I want to do what I do best'
There is also a problem that has not been addressed squarely in the field of mobile AI development. Most Chinese developers are not engineers. They are not good at technological breakthroughs. What they are really good at is operational and business ideas.
However, it is paradoxical that when we talk about AI on mobile phones today, it seems that by default, developers should understand algorithms, understand structures, and understand machine training. Otherwise, it seems that they are not real AI. However, in practice, AI must be done. It is definitely a kind of technical prejudice that the nation understands the algorithm.
As an instrumentalized back-end technology, AI is obviously a more reasonable way to develop the application, apply the application, integrate the efficient task allocation under the same platform, and it is possible to have a more reasonable ecological.
Back to the developer side, I think most of the mobile developers need AI to do not need to start learning complex algorithms and models, but to know where to access these algorithms, directly to my app, and Let me know clearly what more cool things can be done next.
All in all, what developers really should do is to create their own creativity and business insight. Technology should be more and more friendly and simple, instead of making every developer a full-link expert.
The goal of the HiAI engine is to enable developers to develop high-quality AI applications without knowing AI algorithms through the open application layer API. The focus is entirely on the application experience and business practices. This requires the platform to have multiple capabilities. Fully open to developers, open AI open basic environments at different levels, so that developers with different needs and different foundations can effectively select the appropriate method. Even if developers do not understand the algorithm, they can use their own APP for a short period of time. Targeted AI.
In line with this condition, the current world only has the HiAI framework itself. Apple and Samsung currently have adopted closed policies on the capabilities of AI chips, not to mention the comprehensive opening.
The most prominent industry value of HiAI is obviously that it provides different levels of domain-level capability output at the platform level, returning the option to the developer side.
Through instrumentalized, fully-assisted AI architecture support, allowing developers to return to business and creativity, perhaps the right path for AI, is also the future of AI's ecology.
What we may be looking forward to hereafter is that developers will continue to learn and explore new environments and new foundations to quickly create phenomenal AI mobile applications that can arouse widespread attention.
When flowers and the stage are all ready, it's probably about the developer's own.