Posted by Rob Knies

TechFest logo

Welcome to TechFest 2013, Microsoft Research’s annual tech showcase featuring many new, fascinating projects that point toward the future of computing.

Over the next couple of days, we’ll be shining a light on some of the high-profile projects being shown during the event, held at the Microsoft corporate headquarters, in Redmond, Wash.

Just after the event began, for example, I got a chance to chat with John Bronskill, partner architect at Microsoft Research Cambridge, about an intriguing effort called Adaptive Machine Learning for Real-Time Streaming.

As I listened to his pitch, though, it crossed my mind that Bronskill, at 9:30 in the morning, might have been just a bit … hungry.

The project, a collaboration between Microsoft Research Cambridge and Advanced Technology Labs Europe, based in Aachen, Germany, addresses the need to monitor real-time data to determine when something goes awry in a manufacturing environment.

“The idea is that it takes in real-time data and does machine learning to understand when something is not right,” Bronskill explains. “It’s combining real-time stream processing and machine learning, and raising alerts or proposing actions.”

He invoked a food-related analogy.

“The [scenario] we’re concentrating on is a silicon-wafer manufacturing process,” he says. “The silicon is in a tube, like salami: A crystalline structure grows, and then they slice it, literally like salami, so you get this circular wafer. Then integrated circuits are imprinted on it through a series of manufacturing steps. These wafers are very, very expensive, just material-wise, and it can take hours to make a chip.

“If any of the machines or the processes along the line is out of whack, you’re wasting a lot of materials, and you’re going to get a defective part out of it. Instead of waiting till the end, when you say, “Oh, it didn’t work,” if we could monitor some sensors from the machines along the way, you could stop the process a lot sooner, save materials, and detect the issues sooner.”

Get Microsoft Silverlight

As it turns out, the combination of machine learning and real-time data streaming is a great recipe for Microsoft.

“It’s a bit like chocolate and peanut butter,” Bronskill smiles. “Machine learning is a well-established area, and real-time sensor data on a shop floor is nothing unusual. It’s the idea of putting machine learning and real-time data streaming together to make smart monitoring work. That’s the novel part.

“For Microsoft, this is a great thing, because we’ve got some great machine-learning technology, but also, we have this streaming database technology called StreamInsight. It’s a nice, strategic position for Microsoft, where two disparate things are put together and there’s a lot of labor saving.

“The idea is that it knows what abnormal is and raises an alarm. It’s basically automating that process.”