MS Explains Machine Learning Role in Windows 10 Update Experience
Thanks, Mr. Brinkmann! He posted a Ghacks.net item yesterday that alerted me to a fascinating Microsoft Tech Community blog post dated September 26. Entitled “Using machine learning to improve the Windows 10 update experience.” It comes from a pair of MS data scientists named Archana Ramesh and Michael Stephenson. Their story basically explains that MS monitors up to 35 “areas of PC health.” (Each is a collection of devices and/or facilities that could be subject to driver or installation issues.) Machine Learning (ML) lets the company use this data to offer updates to some PCs, but not others. This graph compares uninstalls, crashes and driver issues for ML-selected systems versus the whole population (baseline):
Note: for Kernel Mode Crashes and Driver Issues, the ML model selected PCs experience fewer problems than the baseline.
[Click image for full-sized view.]
Why MS Explains Machine Learning Role in Windows 10 Update Experience?
Basically, MS gathers information about Windows PCs and their configurations. This data gets crunched so machine learning can identify PCs for which updates work. This data comes from the Insider program, or others who voluntarily opt to install certain updates. Those configurations that work well are noted, for inclusion in the update offer. MS also recognizes likely problem PCs. Their receipt of an update offer becomes contingent on a fix for one or more blocking issues becoming available. Only when fixes are in place, is an update offer extended. The following flow chart visualizes this process:a>
Machines whose attributes indicate they’ll have no problems get the update offer. Those likely to face issues don’t come into play until fixes for expected issues are available.
[Click image for full-sized view.]
To me, this provides yet another good reason why Windows 10 users should permit MS to grab telemetry info from their PCs. Though certain suspicious individuals routinely silence their Windows 10 PCs and prevent them from “phoning home” to Microsoft, this kind of data is absolutely invaluable. And indeed, in my own experience, MS continues to improve in its detection of PCs likely to succeed with specific updates (and those likely to hit snags as well).
The authors also provide one of the best explanations of so-called safeguard holds I’ve seen anywhere from Microsoft. Look for a section entitled “Identifying safeguard holds” and read it over. It shows exactly how ML recognizes PCs that need them, and can make a big difference for users and IT pros alike.