spend more than what was previously budgeted, but that’s about it. “It’s a good time to be in Microsoft. On the outside, some companies need to fight for survival. Some companies need to go back to basics—most companies,” he says. “For me, the best thing is I get to continue working on what I was working on a year ago, six months ago—do not change course. And have confidence in the world that things will be getting better. It’s just a matter of time.”
Without further ado, here are some of the best demos I saw, projects that illustrate not only hot topics like energy conservation and search, but the global nature of Microsoft’s research efforts:
—Image-based advertising. Matt Scott from Microsoft Research Asia presented software that lets online advertisers bid on images instead of keywords. The idea is that a user could be shown ads from a restaurant if he or she browses a cooking magazine cover or a street photo of an eatery, say. Object-recognition algorithms (and computers’ understanding of images) are just becoming accurate enough to make this doable.
—Stitching mobile videos (left). You’ve probably heard of software that stitches photos together to make larger composite images and virtual environments (like Microsoft’s Photosynth). Now Ayman Kaheel and his team at Microsoft’s innovation center in Cairo, Egypt, have developed software that combines video streams captured from different cell phones into one higher-resolution video, all on the fly. It could potentially work for events shot from different angles, or for scenes of emergencies stitched together to aid rescuers.
—Web search summaries. Wei-Ying Ma and Zaiqing Nie of Microsoft Research Asia presented a new Internet search feature that represents Microsoft’s increased efforts in Web mining—digging up deeper connections between entities on Web pages so as to produce more useful search results. Their software, called Renlifang, creates a social-network graph for each person or entity you query, showing its connections to other people and organizations on the Web. It reminds me a bit of what Seattle startup Evri is doing to understan
d the connections between Web pages.
—Low-power data centers (left). Janine Harrison and Jim Larus of Microsoft Research in Redmond presented an experimental prototype of a server running on energy-efficient netbook processors that give roughly 50 percent of the performance of the high-power processors currently used in Microsoft data centers—but at roughly 25 percent of the cost, and 15 percent of the power. The key to being able to handle high volumes of search queries with more limited hardware is algorithms that continuously predict the demand cycles so that processors can be switched off when they’re not needed.
