Innovation in D.C. Limelight
June 24, 2009 5:00 PM PT

In troubled times, we often find solace in the fervent belief that tomorrow will be better than today. Millennia of experience have taught us that progress, be it halting or transformational, is natural, inevitable. Such expectations enable us to press forward despite the momentary setbacks we might encounter along the way.

Nowhere, of course, is such inexorable innovation better observed than in the world of ongoing technological advances, in which it seems that today’s marvel is only eclipsed by that of tomorrow. Such promise was the theme of Microsoft Research’s D.C. TechFair 2009, held June 24 at the Newseum, in Washington, D.C., and co-hosted by Rick Rashid, senior vice president of Microsoft Research, and Fred Humphries, managing director of U.S. Government Affairs for Microsoft.

A stream of policymakers, customers, and partners visited the venue to get a glimpse of a tomorrow just around the corner, one that will deliver wonders to computer users worldwide. A total of 13 forward-looking demonstrations offered attendees a look at new, creative ways to use computing to solve real-world problems.

“We believe it is critical, for not only companies like Microsoft, but also the federal government, to continue investing in basic research,” Rashid said in a blog posting. “Some of the technologies we’re demonstrating may have a real-world application today, while others will have an application two, three, or even 10 years from now.”

The demos on display during the D.C. TechFair represent work being pursued by Microsoft Research facilities around the globe―and that representing collaboration with the world’s top universities. The projects featured groundbreaking efforts in computer graphics, online search, workplace collaboration, electronic health records, consumer privacy, and more.

In addition, the work witnessed by the throngs who poured into the Newseum on a sunny, warm summer day In the U.S. capital provided testament to Microsoft Research’s commitment to both basic and applied research in computer science and software engineering. A few of the more compelling demonstrations:

Research in the Round

One of the most visually striking of the exhibits was called Interactions with an Omni-Directional Projector, presented by Andy Wilson, a senior researcher in the Adaptive Systems and Interaction group at Microsoft Research Redmond, and Hrvoje Benko, a fellow researcher in the same group. Standing in front of a seven-foot, geodesic-dome-topped structure during one of the few moments when he wasn’t deluged by inquiring visitors, Benko took a few minutes to explain.

“What Andy and I are exploring here is a new way of interacting with freehand gestures in mid-air,” Benko said. “Most of our research is in human-computer interactions from the perspective of gestural interfaces―’natural’ user interfaces, a lot of people call them these days. How do you move away from standard kind of controls―buttons, mice, and keyboards? Those are very useful for certain kinds of applications. But how do you get to the next level of interactivity, which has been demonstrated with all sorts of products right now that are relying on touch and contact?”

Wilson and Benko are examining projects “above the surface,” enabling people to use their hands to manipulate data. As part of that effort, they are looking at omni-directional data and projecting it in a dome, using imagery from the WorldWide Telescope to demonstrate new modes of interaction. The work is exploring a number of different avenues.

“We’re looking at new ways of interactivity, new ways of gesturing,” Benko said. “We’ve been playing with depth-sensing cameras for a while now. How do you use gesture as an interface that is a little bit further away? How do you employ certain things right in front of the display, using both touch and hover space above the display to do some interesting 3-D interactions?”

The dome projections also enable investigations into the use of curved displays.

“We’re working on a large, immersive experience,” he added. “How does that change the interaction? What are some of the inherent benefits of dealing with a curved projection? Is it better for visualization of certain kind of data? It’s interesting to think of a new environment, something not necessarily mapped to a flat display.”

And Benko and Wilson aren’t stopping there.

“What if, instead of a regular projector in a room, you have one of these?” Benko asks. “Not only can you project on every surface in the room, but you can make them interactive. You can move things between surfaces. All of a sudden, the whole game changes. It’s not like you have one wall of a conference room being your ‘PowerPoint wall.’ All of a sudden, you have this gestural system that lets you move data all the way around, between all these different surfaces that can potentially be interactive.”

Better Recommendations

Adam Kalai of Microsoft Research New England showed a demo called Recommendation Systems that attempts to improve upon the Web practice of users making recommendations about products or services.

“This demo is about coming up with a good recommendation system to figure out a good way to give personalized recommendations to people when you have a network, for example a social network, where people show who their friends are,” Kalai said. “We have other networks where people express who they trust. We have a particular way of evaluating a recommendation system based on the properties we want it to have.

“A good recommendation system should have a property that, if you trust somebody directly, that should be more important than if you trust them indirectly through a friend. On the other hand, we want to take into account these indirect trust relationships. We also have an interesting property where it’s resistant to spam, so a spammer can’t mess with you too much by pretending he has a million friends who all trust him.”

In this instance, Kalai’s work appears to be on the cutting edge of current explorations.

“I don’t think there’s been very much work in this area,” he says. “There’s been some, but I think we get very, very good algorithms because we come from this perspective of deciding what are the right properties of an algorithm to start with.”

It certainly hasn’t been difficult to attract people’s interest.

“We don’t even have to go to look for people,” Kalai smiles. “People have come to us already, a couple of times, where they say: ‘This is exactly what we’re trying to do. We don’t know what algorithm to use. Can you make a suggestion?’ “

Resources in One Place

Natasa Milic-Frayling, director of Research Partnership for Microsoft Research Cambridge, is well aware of the challenges encountered by researchers of all stripes, and she’s doing something about it.

During the D.C. TechFair, she showed Research Desktop Activity, designed to enable the collection of all resources related to a particular project.

“What this is solving,” Milic-Frayling says, “is the problem of us gathering, then re-accessing, information. We lose lots of time doing that.

“Research Desktop is a collection of facilities that help us focus on work, on the creative part, meaning that we can easily collect information so we can conduct our research. That’s done by associating tags with objects and resources we need to do our work. For example, we have support for what we call activity management. If I have access to information on the Web, I have things in my e-mail, and I have documents in my file system, I can associate them by typing in the tag, which automatically propagates in all these instances. Tomorrow, the day after tomorrow, or three weeks from now, I may come back. I just click on that tag, and it lists for me all the resources that I need to continue my work.”

That would free researchers from having to manage separate storage spaces, such as e-mail, file systems, and the Web.

“There’s not much that has been done in activity management in a desktop environment that cuts across all these resources,” Milic-Frayling said. “A lot of tagging has been done online. For example, if you go to Delicious, you can associate tags with photos, but exclusively for photos. Our project is unique because it enables us to collect very heterogeneous resources. In Microsoft Research, we have done work in the past on activities in many different ways. The key here is that the interaction is extremely natural and blends extremely well with managing applications, resources, and tags―all three very well integrated.”

But such work isn’t just for professional researchers. Anybody preparing a study or a paper could benefit from the functionality Research Desktop offers.

”When students are writing a report,” Milic-Fryaling said, “they can tag their document as ‘This is my final report,’ and if they exchange e-mail with their colleagues or professors or go online and collect resources, they can apply the same tag to all of them.

“Therefore, it will support the whole lifecycle, from gathering information, to communication, to analysis, and to writing. That can be done not only by students. It can be done by professors who are collecting information for a class, but it can also be done by market researchers, who have to write a report on a particular aspect of a product or for a particular audience.

She went on to mention that she has a series of videos available. “Six of them,” she pointed out. “They’re all right here.” She then gestured toward the right side of her demo monitor, the portion where Research Desktop displayed them all, in conjunction with other related documents: dynamic research on display.

Botnets Beware

Spammers have been a scourge to e-mail users, who have to wade through a plethora of nonsense to find e-mail they will find actually pertinent. But the tide is turning, thanks to work from Fang Yu, a researcher at Microsoft Research Silicon Valley.

“We detect spammer-created botnet spam e-mail accounts,” Yu explained. “Spammers currently create a large number of spam e-mail accounts from large e-mail providers. There are millions of such accounts, and they are very stealthy, but we are trying to detect these spammer-created accounts to stop them from spamming.”

Yu and colleagues take an interesting approach to thwart such egregious attacks.

“Each individual won’t send a lot of spams,” she explains, “because that would make them easy to identify. Our approach is to look at the correlations between these accounts. You can think of these accounts as all controlled by one person, a spammer. They’ve got to be very similar; they run scripts, so they’re very similar in their actions. We build a graph-based approach to look at the correlations between the accounts. The precise grammars are very similar, and by analyzing the graph, we are able to identify the similarities.”

In this case, the work being done at Microsoft Research is not pointed toward the future. It’s helping to deter spammers in the here and now.

“This is based on real data, real Hotmail Web mail accounts created by botnets,” Yu said. “We’ve captured tens of millions of accounts already, and we are sharing the results with product groups. We also have compromised hosts’ IP addresses to see where these are coming from.”

Someday soon, perhaps, users will be able to open their mailboxes without cringing at what they find.

The Importance of Basic Research

In his pre-event blog posting, Rashid underscored the need to continue to foster innovation.

“Our goal,” he wrote “is to advance the frontiers of computer science, not just Microsoft's interests. Our investment in basic research creates value not just for a company like Microsoft, but for society at large.

“Such investment allows companies, countries, and whole societies to respond rapidly to change. It’s this agility that is essential.”