Huawei MindSpore (the platform for development of AI)

2020: Disclosure of source codes

At the end of March, 2020 Huawei announced disclosure of source codes of the MindSpore environment  allowing to develop applications with algorithms of artificial intelligence. Source codes are laid out on the GitHub and Gitee portals.

According to the VentureBeat portal, the facilitated MindSpore platform TensorFlow from Google and PyTorch from Facebook is similar it is scaled on devices, in peripheral and cloud environments, helping developers to master quicker creation processes AI software.

Structure of the MindSpore environment

By the end of March, 2020 Huawei got support of a number of the partners helping to develop MindSpore. Among them — the University of Edinborough (University of Edinborough), Peking University (Peking University), Imperial College London and a robotic startup of Milvus.

Developers of a framework state that the tool is aimed at saving privacy of users. MindSpore uses only cuts and data models instead of work with data directly.^[1]

Thanks to what in typical neural networks for NLP (Natural Language Processing) of MindSpore requires less lines of the master code, than other platforms, efficiency of development as the producer claims the minimum increases by 50%.

At the same time infrastructure of Huawei MindSpore supports not only own nayroprotsessor like the mentioned Ascend 910, but also other processors and graphic accelerators, available in the market.

One more advantage of the platform is that it does not process  data therefore confidential information of the user is effectively protected even when using in cross-scenarios. Besides, MindSpore has the built-in technology of protection of models to guarantee their reliability and security, declare in Huawei.^[2]

2019: Announcement

At the end of August, 2019 Huawei provided the open environment for development of AI-applications of MindSpore. The company claims that this platform contains 20% less lines of the master code in comparison with similar solutions of the leading developers.

According to the statement of Huawei, the Mindspore environment was created for joint work with the Ascend 910 processor thanks to which learning process of AI models proceeds twice quicker in comparison with similar solutions in which the program library TensorFlow is used.

Huawei provided the open environment for development of AI-applications

In addition to Ascend chips, MindSpore also supports graphic, central and other types of processors. 

According to the company, computing platforms for artificial intelligence have crucial value for a year of simplification of development of AI-applications, their distribution, increase in their availability and security.

In 2018 Huawei designated three criteria to which the developed AI platform should answer:

• Simplicity of development — it significantly reduces duration and training cost.
• Efficiency — the smallest quantity of resources with the greatest possible energy efficiency should be used.
• Adaptability to any scenarios of use — including devices, peripheral and cloud computing.

MindSpore is one of milestones in achievement of these purposes. As privacy protection purchases the increasing value, necessary is a support of all scenarios of use for ensuring safe and widespread artificial intelligence. It is a crucial element in structure of MindSpore therefore the platform can be adapted to different conditions of deployment. MindSpore will be suitable for any environments — big and small — regardless of the budget of resources.

MindSpore helps to provide confidentiality of the user as deals only with variable and model information which was already processed. The platform does not process data therefore confidential information of the user is effectively protected even when using in cross-scenarios. Besides, MindSpore has the built-in technology of protection of models to guarantee their reliability and security.

The MindSpore AI platform adapts to any scenarios of use with any devices, peripheral and cloud environments and also provides interaction between them on demand. Its basic concept "AI algorithm as the code" allows developers to develop AI-applications and quicker to train the models.

In normal neural network for natural languag processing (NLP) MindSpore uses for August, 2019 20% less lines of the master code, than the leading platforms in the market.

MindSpore helps to provide confidentiality of the user as deals only with variable and model information which was already processed. The platform does not process data therefore confidential information of the user is effectively protected even when using in cross-scenarios. Besides, MindSpore has the built-in technology of protection of models to guarantee their reliability and security, the press release says.

The Chinese company also says that the new solution allows developers to perform the difficult calculations connected with artificial intelligence and satisfies the need for various computing powers for different scopes of application.

MindSpore supports different platforms, including peripheral and cloud computing and also devices. The solution development which allows developers easier advances the concept and quicker to train the models, ZDNet tells.

Emergence of MindSpore in a public access is planned on the first quarter 2020. This product will become a part of strategy of Huawei aimed at the accelerated development and implementation of AI technologies.^[3]

Robotics

Notes