Shulou(Shulou.com)06/02 Report--

Today, more and more people have begun to use the power of AI to change their production and life.

They may be professional deep learning algorithm engineers who can use large-scale computing power and data sets to create unprecedented AI capabilities; they may also be a plant research scientist who hopes to complete the classification and popularization of thousands of plant varieties with AI; or they may be a student who is eager to explore artificial intelligence.

AI, as a technology engine, is relying on the creativity of various fields and talents in an unprecedented way. If you also believe in AI and are hungry for technology, and have had the idea of "giving AI to others" in your mind, you may have the potential to become a "MindSpore Evangelist."

We know that MindSpore is a deep learning training reasoning framework built on the basis of Shengteng AI processor. Its main audience is data scientists, algorithm engineers and other groups, helping them implement AI functions such as computer vision and natural language processing.

And what the hell is this preacher? Is it the legendary "bring your own dry food" and generate electricity with love? What are the similarities and differences between developers?

Simply put, from developers to evangelists, it is not only a leap of faith in AI platforms, but also a twist between technology companies and technical talents.

From collaborator to lover: the leap of faith between evangelist and technology platform

The "preachers" in the public impression often have both professional skills and "literary skills." So for the majority of AI developers and engineers, what does it mean to become a platform "evangelist"?

First of all, the premise of "preaching" is that programmers themselves have a strong foundation of trust and love for the platform and technology itself, so that they can have motivation and enthusiasm to close the distance between the public and technology.

Secondly, the existence of "evangelist" can publicize technology to developers based on code, make various technical means provided by the platform easier to understand and apply, fully release the possibility of technology at the individual end, and open up the "extraordinary meridians" of developers. This requires that the persuasiveness of his personal brand should be reliable enough. After all, a pancake fruit owner enthusiastically promotes hot dry noodles, which must also make people suspicious. Google's chief evangelist, Vint Cerf, the father of the Internet, and Microsoft's former evangelist, Vic Gundotra, demonstrate the latest technology at every developer conference.

In addition, the platform's technology strategy, development audience, and technology zenith determine the original intention, strategy, and method of preaching, and how the evangelist persuades the target audience to accept the platform's solution. This process obviously requires continuous efforts to gradually take root in industrial practice and translate into long-term trust that other technicians can feel.

It is not difficult to find that technical evangelists not only need the blood and ideal of a cavity pioneer, but also need the platform to have long-term, reliable and strong backing, so as not to be "dragged down" and become "ten years of drinking ice."

Therefore, technology evangelists and technology enterprises are not only the cooperative relationship between developers and platforms, but also emotional partners who practice technology ideals and dissemination from the heart.

So, MindSpore recruits preachers, what are the reasons and opportunities?

MindSpore re-upgrade, China's full-scene AI development framework "seek journey"

Before we solve the mystery of MindSpore Evangelist, perhaps we should first solve a basic problem, which is-He Tao Ke Bu.

We know that the AI field is a Shura field for talent. According to the latest survey by recruitment website Indeed, machine learning engineers, data scientists, computer vision engineers, database builders and algorithm engineers are at the top of the list.

Such a group of top-notch professionals, obviously rely solely on money to drive is not enough for them to easily deliver trust. And gathering high-quality developers is inseparable from the benchmarking role of these leaders.

Therefore, in the matter of deep learning development platform, the core of competition naturally falls on the recognition of technical ability, resonance of technical ideal and practice of technical value.

So, what is so special about MindSpore's approach?

First of all, the technical ability is second to none.

At present, there are many frameworks and platforms that can provide machine learning and deep learning training in the world, and the mainstream framework is more than one hand can count. The primary consideration for developers to choose a platform for AI development is whether the technical capabilities of the platform are reliable and easy to use. This determines whether excessive learning costs will be consumed and whether the community can provide sufficient strong resources to support it.

As we know, MindSpore, as a full-scene AI framework, can be said to reflect an incomparable temperament in lowering the threshold for AI application development.

Deep learning frameworks are not easy to mix without functional innovation, and MindSpore is launched after popular frameworks such as TensorFlow and PyTorch, which can still attract many developers. Obviously, there are also two brushes.

MindSpore's first open source version is version 0.1.0-alpha, which can provide native support and software and hardware co-optimization with Shengteng AI processor. The data shows that "Shengteng 910" combined with MindSpore shows nearly twice the performance improvement compared with the existing mainstream training single card combined with TensorFlow. This allows developers to unleash AI imagination with sufficient selectivity and computing power.

In addition, with the help of industry-leading technologies such as Source-to-Source automatic differentiation and distributed parallel training, combined with complete data preprocessing modules, efficient graph calculation engines, visualization tools, model evaluation tools and other features and components, MindSpore's development, operation and deployment are also efficient and friendly.

And recently completed an iteration of the new version of 0.2.0-alpha, and added support for Windows 10 system, and Linux CPU version is no different, so that more small partners using Windows can also enter the AI development ecosystem. In addition, MindSpore version 0.2 released Benchmark, which includes ResNet-50 v1.5, BERT-Large and other models commonly used in machine vision, and the training speed on Ascending AI processor has been significantly upgraded. For example, image training processes 27090 images per second in 16*Ascend 910 scenes, natural language processing processes 1613 sentences per second in 8*Ascend 910 scenes, and the acceleration ratio of multi-machine parallel training reaches about 0.95, and the development efficiency is greatly improved.

On the basis of a generation of visualization tools, MindSpore 0.2 version further introduced a new feature "parameter distribution map," which can observe the weight and gradient changes with the training process during the network optimization process and judge the optimization progress of the network. Model traceability draws many super-parameter configurations such as learning rate, loss value and model size together, which is convenient for developers to adjust parameters.

More importantly, MindSpore provides end-to-end, edge, cloud independent/collaborative development capabilities, supports large and small, and adapts to independent deployment of all scenarios. This means that developers can complete application development on a platform, and can enjoy a consistent development and debugging experience in all terminal device deployments, which is also very attractive for the future of AIoT pan-intelligent terminals.

In any case, developers think of, unexpected, MindSpore in the continuous evolution of more and more, I heard that at the end of May will also send v0.3.0-alpha. The iteration rate of one change per month, do you MindSpore engineers still have hair?

Secondly, self-developed Chinese framework.

Today, we talk about independent deep learning open source frameworks, and everyone may be a little tired of listening. However, with the uncertainty of the global situation in the post-epidemic period, it is clear that Chinese AI developers will put the lifeline of future growth on the foundation of the Chinese platform, at least at the psychological level.

On the other hand, with the pace of China's new infrastructure, it is an inevitable fact that AI sinks to thousands of industries. However, at the development ecological level, community prosperity is still high and far away. With more industrial forces such as MindSpore exported, it is natural to establish a more autonomous and vocal Chinese AI development system from the source.

While accumulating unique AI development foundations and data sets, flexible development and deployment are carried out according to the special needs of vertical industries, which are obviously the "near water towers" that MindSpore can provide. At the same time, it also provides another solution for AI to enter the social core fields such as security, city and government affairs.

Finally, although it is difficult to industrialize AI.

Then, after completing the high-level Chinese deep learning framework and ecological construction, is it done? There is such a "paragraph" circulating in the AI world that AI falls into a situation where big cattle dig pits, calves fill pits, engineers adjust their references, and laymen doubt.

MindSpore, which adheres to the spirit of sharing and openness, has practiced the cultural core of open source community through community organizations such as special interest group SIG and technical governance committee since its establishment, combining community autonomy with expert open governance to help ordinary developers grow rapidly and promote AI penetration into various fields. The participation of Evangelist Program will also strongly promote the free sharing of open source community and the fertile ground for technology sharing.

How to make the majority of ordinary people from "doubt" to conviction requires not only the efforts of evangelists and technology enterprises, but also the key power point of technology landing. Especially at the moment when algorithm innovation enters the bottleneck period, it may be the proper meaning of an open source framework to continuously deepen the industrial sinking of AI and assist the intelligent transformation of all walks of life.

We know that unlike the traditional route of European and American frameworks from laboratory to industry, MindSpore is a deep learning framework that is closely integrated with industrial AI development. The muzzle of AI is directly aimed at the pain points and problems of industrial application technology. Therefore, under the comprehensive offensive of Shengteng and MindSpore, Atlas intelligent computing platform and series of development tools, it has become a powerful catalyst for AI computing power and development throughout the industry from software, hardware and ecology. AI from the laboratory into the city, campus, hospital, factory, park... and so on countless details of the reality of the scene, is the need for such a clear and easy overall plan as a base.

AI is set up on each scene of the edge cloud, which may be MindSpore's AI way, and it is also an era of "collusion" with the evangelists.

The Way of AI, the Heart's Work: Looking for the Wall Breaker and Evangelist of the Times

If MindSpore's ability to recruit evangelists proves itself enough to explain "what is Tao," then it is time to answer another key question-"how to distribute it."

First of all, with the upgrade opportunity of MindSpore 2.0 version, an AI development environment with unified architecture and stronger functions has been formed. From the back-end processor to the inference engine and front-end development tools, it has laid the bridge and premise for the communication between evangelists and AI developers and industry applications.

Secondly, MindSpore also breaks the evangelist system that traditional technology companies rely on Big Bull Aura. As long as MindSpore developers can apply. In addition, incentives are designed in accordance with the pyramid structure of evangelists and community contributions. On the one hand, professional scientists and researchers can be attracted to establish a top-level AI content sharing ecosystem, and developers with vertical field development experience can share their experiences and experiences with passion, thus accumulating richer community resources for AI to blossom and combine more quickly in the industry and deploy efficiently.

When the ripples of the industry of "MindSpore+ Evangelist +AI Developer + Thousands of Industries" continue to spread, the lake surface of the whole intelligent society will also constantly surge out thousands of ideas and wonderful things hidden under the water. Technicians 'faith and persistence in AI may have another name and honor besides developers from now on.

And this, too, will be a new beginning for our imagination of the intelligent future.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.