As an active player in the field of AI, Microsoft has recently introduced a Windows 10 platform for developers to move their cloud-based AI models over to desktop apps.
Designed specifically for Windows developers, the platform will be made available when the next major version of Windows is launched.
The platform’s new tools will make a typical workday easier for developers as they will soon be able to run machine learning models on their own machines without visiting the cloud, by making use of the GPU on desktops to allow people to run models in real time.
Using the new AI platform, Windows developers can choose to build their AI models within the cloud before integrating them with apps on their local desktop. This allows users to select a framework of choice when developing machine learning models.
According to Kam VedBrat, Microsoft’s Partner Group program manager, the new tool from company “completes the story for Microsoft from an AI perspective”. This new platform is consistent with previous talks from the technology giant about improving developments of infrastructure and tools for AI and machine learning.
A well-known initiative is Onnx, which also makes it easier for developers to move models between different frameworks by allowing the conversion of PyTorch, Caffe2 and CNTK models into the Onnx format. Apart from Microsoft, other technology giants like Amazon and Facebook also support the project.
Via the Azure Custom Vision Service, developers are also able to build image recognition models, which can be exported to and used within Windows ML. Developers using this service are not required to have extensive knowledge on the complicated science behind the building of machine learning models, such as the case of when they are using the more traditional framework. Instead, users are only required to provide the Azure service with tagged training data. These models then use the silicon already available within the user’s machine, which is usually in the form of the CPU or the DirectX 12 graphics card. However, a flexible API will be also made available for users to assess other hardware, such as the future Intel Movidius vision processing units.
According to Kevin Gallo, a Microsoft corporate VP, the new initiatives reduce latency and at the same time improve the level of privacy of users’ information. Another key benefit is a reduction in cost – by lessening the need to run machine learning models within the cloud, the usual costs associated with the traditional approach of running models are reduced.
Following the release of the preview of Visual Studio 15.7, users can enjoy an additional function of easily generating a project’s model interface. Users just need to add an Onnx file to the Universal Windows Platform (UWP) and let Visual Studio do its job. This functionality will also be added to Visual Studio tools for AI.
For valuable advice on Microsoft’s new services and tools, feel free to contact WM Reply to speak to one of our customer service personnel. A friendly team member will be happy to share updated information and provide examples on how other companies and businesses have successfully adopted Microsoft’s products into their organisations.