Microsoft plans to introduce artificial intelligence (AI) features directly into Windows, in a move that could see AI applications more widely used on Windows-powered hardware, particularly mobile and embedded devices.
Microsoft said Windows ML, as the feature is called, will support chips made by Intel-owned Movidius as well as GPUs from companies such as Nvidia and specialty hardware from other manufacturers.
The announcement bolsters Microsoft’s efforts to build up its presence in mobile devices, which are dominated by companies such as Apple and Google.
At its Windows Developer Day event on Wednesday, Microsoft outlined Windows ML and other features set to be included in the next major Windows 10 update, which is due to be finalised this month and to begin rolling out to users in April.
The update, which doesn’t have an official name as yet, is known internally as Redstone 4, taking the name from the US’ second human spaceflight in 1961.
Microsoft said Windows ML is aimed squarely at bringing local AI processing to devices at the edge of networks, which might have tight processing and power constraints.
In the past such devices, including surveillance cameras, drones, mobile phones and other mobile hardware, has depended on passing sensor data to high-powered servers for processing.
A more recent trend is for that processing to take place directly on the device, using hardware accelerators and pre-trained AI models. Huawei, for instance, is one of the first companies to have built an AI chipset specifically for smartphones.
Developers are already taking advantage of these kinds of capabilities, but building them directly into Windows makes it easier for for more people to build edge AI apps.
“For consumers, this is just the beginning,” said Kam VedBrat, group programme manager for the Windows AI Team, in a blog post. “You’ll see more intelligence in everyday experiences on Windows devices.”
He said applications include 3D communications, creating personalised music playlists and organising users’ tasks.
For developers, on-device processing can mean lower latency and lower costs for transferring data across networks, VedBrat said. The platform allows developer to choose whether processing takes place on the device or in the cloud based on the particular scenario.
VedBrat said Windows ML will support AI applications across embedded devices, HoloLens augmented reality hardware, 2-in-1 tablet/laptops such as Surface, desktops, workstations, servers and data centres.
Windows ML is compatible with ONNX, an industry standard for machine learning models that’s also supported by Facebook, Amazon Web Services, Nvidia, Intel, Qualcomm, AMDand others. That means data scientists can train ONNX models and deliver them on Windows ML.
Microsoft said it has built automated ONNX capabilities for developers into Visual Studio Preview 15.7.
One of the companies that stands to benefit from Windows ML is Movidius, a San Mateo-based firm that was bought by Intel in September 2016.
Movidius makes low-power chips specialised for computer vision and AI processing chores that can be built into battery-powered devices such as surveillance cameras and drones.
Founded in 2005 in Dublin, the company’s recent products include a USB stick that can add AI processing capabilities directly to low-power devices and the Myriad X chip, which features an accelerator for neural network-style AIs.
Intel, like Microsoft, has seen its dominance challenged by the rise of mobile devices, which mostly use chips based on low-power ARM designs.
Its purchase of Movidius was a move towards making a bigger impact in the mobile market.
What do you know about artificial intelligence? Try our quiz!
After previously expressing its concern, the British Government now confirms a national security review of…