Microsoft is giving Windows developers another reason to look forward to this spring's Windows 10 feature update, code-named Redstone 4.
The Redmond, Wash., software giant announced on March 7 that the next major system software update will introduce Windows ML, a set of new components that will enable coders to imbue their apps with artificial intelligence (AI) capabilities. "We are now going to allow high-performance ML [machine-learning] model evaluation on Windows," Microsoft Corporate Vice President Kevin Gallo told eWEEK.
Developers will soon be able to tap into the computing resources of their PCs and edge devices, enabling hardware-accelerated AI models in Windows across a wide spectrum of devices with varying hardware configurations, Gallo explained. The capability will enable millions of .NET and Windows developers to easily incorporate AI techniques and functionality into their applications, he added.
Why evaluate machine-learning models on a local PC rather than AI-enabled cloud products, including Microsoft's own Azure platform?
Although the cloud does enable large-scale AI workloads to tap into the massive compute resources of hyperscale data centers, there are some crucial benefits to taking a client-based approach, explained Gallo. AI evaluation conducted on a local system "reduces latency and gives you real-time results," he said. Users will be able to run analysis on large volumes of video, images and other local data by harnessing the processing capabilities of a Windows desktop or other device.
Additionally, it reduces cost in the monetary sense as well as in bandwidth and compute time by exploiting the CPU or GPU of a local machine. Finally, Gallo has observed an increased need for organizations to keep data used in AI processing within the confines of their own networks, whether due to the outsized effort and cost of migrating large volumes of data to the cloud, compliance requirements or other factors.
In terms of the developer experience, Kam VedBrat, a Microsoft partner group program manager, said the new capabilities are accessible through "a new API surface" that is easy to target and integrates seamlessly into the software development workflow.
Coders with problems that are hard to solve using traditional algorithms will be able to take a pre-trained machine-learning model, and have it show up "as a first-class object" in Visual Studio, which developers can then use to add AI features to their applications. A manufacturer, for example, can use an image classifier based on existing photos of good parts and broken parts, taking the resulting model and using it to build applications that can tell the difference, VedBrat said.
In the realm of IT planning and management, businesses can use server logs and practically any other types of information gathered by their systems to inform AI-enabled applications that evaluate past patterns and help users make smarter technology decisions.
Additionally, "Windows 10 is going to support ONNX," the open-source format for AI framework interoperability originally co-developed by Microsoft and Facebook, VedBrat revealed. In the preview of Visual Studio 15.7, users will be able to add ONNX files to UWP (Universal Windows Platform) projects, and the IDE will automatically create the necessary model interfaces.