Microsoft says its machine-learning models for selecting which devices are ready to upgrade trouble-free are delivering major results, slashing the number of crashes users experience after upgrading to a new version of Windows 10.
As two Microsoft data scientists explain in a blogpost, the Windows 10 version 1803 rollout was the first time Microsoft had used machine learning (ML) to release Windows 10 to the world.
It was the fastest-ever adoption rate for Windows 10 and, after it reached 250 million installs, Microsoft boasted how AI had helped it “go faster” with Windows 10 rollouts.
SEE: 20 pro tips to make Windows 10 work the way you want (free PDF)
The second time was Windows 10 version 1809, which quickly turned into a disaster, causing deep introspection among those responsible for Windows servicing and delivery about its approach to Windows 10 rollouts and testing.
Microsoft’s data scientists conspicuously omit any reference to 1809 in the post.
Microsoft also used machine learning in the Windows 10 version 1903 rollout and, thanks to its third use of ML at scale, its models now evaluate “35 areas of PC health” when determining whether specific devices are ready to upgrade and which devices it puts an upgrade block on.
The fourth time will be the Windows 10 version 1909 (19H2) update, which is due out any time now and is a minor update compared with previous September-targeted releases. And if Microsoft is right, users who get the update should experience fewer glitches than in the past.
Microsoft suggests these additional 35 PC health signals have delivered results. It claimed major reductions in stability issues and driver clashes in Windows 10 1803.
Now the researchers say PCs chosen via ML have fewer than half the number of system-initiated uninstalls, half the number of kernel mode crashes, and five times fewer post-update driver issues.
“Machine learning helps us detect potential issues more quickly and helps us decide the best time to update each PC once a new version of Windows is available,” the data scientists write.
The pair also explain how Microsoft is using Windows Insider testers to improve the experience for other users. Once these users update, Microsoft monitors diagnostic data for kernel-mode crashes, system-initiated uninstalls, abnormal shutdowns, driver issues, and other health signals.
ML allows Microsoft to identify issues and potentially implement a block and to protect PCs that haven’t updated yet, allowing Microsoft to contain the problem while it investigates. The model also “predicts and nominates PCs that will have a seamless update experience and should, thus, be offered the update”.
However, a recent run of bugs affecting Insiders and then general users shows this process still misses some things, even when Windows Insiders actually report the bugs.
The blog post is more focused on the benefits and challenges of building an ML model for assisting with operating system updates. Microsoft used to rely more heavily on humans to test updates before they were released but the company fired significant number of them in 2014 as part of the Nokia purge.