Microsoft sees new Azure tool as a proactive trouble-shooter

Microsoft envisions a future where systems can predict malfunctions in devices and buildings before they occur. First step: a cloud-based machine-learning tool that goes into public beta in July.

Charles Cooper Former Executive Editor / News
Charles Cooper was an executive editor at CNET News. He has covered technology and business for more than 25 years, working at CBSNews.com, the Associated Press, Computer & Software News, Computer Shopper, PC Week, and ZDNet.
Charles Cooper
2 min read

Getty Images

Satya Nadella's first tweet came in mid-February 2009 and it consisted of two words.

"Machine learning!" he wrote.

Even for Twitter's 140 character limit, that was a brief entry and was more cryptic than the usual inaugural tweet -- something along the lines of "Knicks stink" or "Springsteen rocks" -- but in retrospect, it's starting to make sense.

At the time, Nadella, who became Microsoft's chief executive this past February, was in charge of research and development for the company's Online Services division. It turns out that Microsoft was embedding this kind of predictive machine-learning technology into its Bing search engine to return relevant answers to user queries in ranked order.

Fast forward to July: that's when Microsoft releases a beta version of its technology to anyone signing into its Azure cloud computing service. Microsoft says this will dramatically reduce development times to hours, sometimes minutes, thus simplifying a process that sometimes would require months to build machine learning models, according to Joseph Sirosh, the newly appointed vice president of machine learning at Microsoft.

"This brings all the benefits of cloud computing to machine learning," he said. "That's a game changer."

The pitch is that this sort of cloud-based offering democratizes what has been an esoteric field and will dramatically improve a company's forecasting ability.

Microsoft is betting that as more intelligent devices are built and used, there will be a commensurate increase in the amount of data residing in the cloud. Then it's up to companies to build tools on top of that data which can garner that data for real-world use.

Sirosh used the example of elevators with built-in sensors, which would allow maintenance engineers to monitor them and predict when malfunctions were likely to occur and head them off. The same sort of predictive maintenance would apply to automobile assembly lines where sensors could identify flaws in the cars very early in the process and obviate the need for expensive and embarrassing recalls. But first you need to build the software systems that can do the job, and that's where this latest reason to use Azure comes in.

He said that in the past data scientists who composed these sorts of machine learning models handed them over to IT to build what were highly sophisticated systems, but a lot got lost in translation. "And now there's nothing lost in translation," according to Sirosh. "Every engineer, every applied mathematician -- everyone can become a very effective data scientist without having to learn an incredibly sophisticated cloud program."

Carnegie Mellon has been experimenting with the technology to collect sensor data from different buildings to measure water and energy use. "Within a couple of hours, they were able to connect streaming data that previously would have taken days and weeks, and now they could do it within hours for fault detection and diagnosis," said Prabal Acharyya, of OSIsoft, an application developer who worked on this project with Microsoft. "It's a leapfrog development."