Microsoft Research Brings Breakthroughs to Bottom Line

Commentary: Pure research yields unexpected applications when scientists, product teams talk and trust.

Im still assimilating pages of notes, and hours of recordings, from two full days at Microsoft last week. Youll see some of what we learned there, including comments from almost an hour of Q&A with Steve Ballmer, in eWEEK on Nov. 18--but right now, Id like to share some things that probably wont be featured in that coverage.

Of more than a dozen discussions, the one that most fits the mission of these letters may have been our hour with Rick Rashid, senior VP of Microsoft Research. Rashid commands my attention because his labs continually demonstrate that a research group can do it all: produce publishable work on the fringe of science, and also contribute to near-term product improvement.

"The first goal we have is creating new science, moving the state of the art forward," said Rashid, adding that Microsoft contributes a disproportionate number of the papers at many professional conferences: In graphics, for example, he estimated, "weve had more papers published over the last six years than any [other] two or three organizations combined; just our China labs publish more papers than almost any other organization."

Assessment through the peer review process of conferences and journals, he said, is "the only way for a corporate research lab to make sure that its not fooling itself in thinking its doing good stuff. Youve got a lot of outside eyes looking at what youre doing."

But if Microsoft Research isnt keeping itself on a short leash, and taking its mandate from the product development groups as to where it will focus its attention, then why has it been so successful in bringing its results to market in the form of significant product improvements? Microsoft seems to be better than most at cultivating "demand-side" awareness, in the product groups, of the possibility that pure research may yield unexpectedly useful results.

"Microsofts a very receptive organization, and we have a number of systematic things we do: I have a team of program managers whose sole job is to manage interactions between product groups and research groups, to shepherd technology transfer," Rashid explained. He emphasized the importance of personal relationships across group boundaries: "Ideas flow because people trust each other, not because people are just dumbstruck by the great idea. You develop a track record of creating good things, and people come to believe that putting your ideas into their products will give them better products."

Rashid isnt about to get into the business of trying to guide research based on preconceived ideas of what it would be useful to discover. "The things you really dont expect to have an impact, all of a sudden have an impact," he said. "Extremely theoretical work turns out to be important to the Windows team: An algorithm turns out to be applicable to networks, or scheduling.

"Support vector machines, a very off-the-beaten-path thing from the graphics world, relate to drawing the best plane to divide two groups of things in a multidimensional space: It was always extremely slow, but some of our teams thought they might be able to use it and developed some revolutionary techniques to speed up that technique for real-time classification. Its found applications all through our products for handwriting recognition, speech recognition, its used in our Commerce server for managing categorization of things that consumers might want based on what other peoples interests have been in the past.

"All of a sudden, something thats been around for a long time becomes widely used because of a theoretical breakthrough in the way its implemented."

When the people on the leading edge of science are in regular, collegial contact with the people on the grinding edge of bringing products to market, they want to help each other: "We dont do anything specifically to make what the researchers do be relevant to our products, other than to put people in contact with each other, so that when one of them has a problem, the other may be able to respond, Oh, I know what you can do about that," said Rashid. Doesnt that sound like the kind of conversation youd like to hear in your halls?

Tell me how IT R&D needs to deliver the future today.