Summary:

Microsoft Research distinguished scientist and manager John Platt explains the impact of machine learning on current and future Microsoft technologies.

John Platt, Distinguished Scientist and Manager, Machine Learnings Department

When you think about Microsoft, you might not think about machine learning. But you can find it pretty much everywhere, and it plays a much bigger role than you’d realize. John Platt, distinguished scientist and manager in the machine learning department of Microsoft’s Research arm, explained the important of machine learning for Microsoft at Gigaom’s Structure Data conference.

“Machine learning is actually pervasive throughout pretty much all Microsoft products,” Platt said. “For example, in speech recognition, there is a standard benchmark called switchboard, and it was stuck for ten years. There was no improvement, even though the data set was getting much larger. And then some of the folks from Microsoft Research came in started applying these deep learning techniques and actually got initial reductions of 30 percent in error rate.”

You can also see machine learning in action in technologies you’re more likely to use everyday. “Whenever you use Bing,” Platt said, “You’re using many, many components that are trained with machine learning… The only way you can answer however many questions are being [asked] is to have something that operates independently, autonomously.”

Microsoft’s Xbox uses machine learning to a great extent as well. “The fact that it can see you in the room even though it’s poor lighting and you can actually wave your arms and it can track you, that’s all done with a machine learning piece of software,” Platt explained.

As for the future, Platt spoke about the use of machine learning to improve security. He explained that it allows analysts to search through lots of data more effectively, or to power code that autonomously finds malware authors on its own.

A video embed of the session follows below:

Photo courtesy Jakub Mosur.

Structure Data Ticker

Comments have been disabled for this post