Microsoft Research New England Turns 3: Jennifer Chayes Reveals Its First Product-and Collaborations With Bing, Facebook, and Twitter
Whether you’re part of the Boston-Cambridge technology community, or just an interested observer, you’ve probably been to the Microsoft New England Research and Development Center (known affectionately as NERD) at least a few times in the past year. But do you know what Microsoft is actually working on there? I didn’t.
Until last week, that is, when I visited Microsoft Research New England, one of the main groups at NERD, housed near Kendall Square. The tower at One Memorial Drive rises majestically from the banks of the Charles River, but it guards its secrets closely. Jennifer Chayes, the managing director of Microsoft Research New England, which occupies the 12th floor, reveals the projects her group is working on slowly. But she does reveal them.
OK, it’s an open research lab with hundreds of visitors and collaborators from academia and industry, so secrecy isn’t the culture here—at least, not amongst the research community. But from the lab’s opening in 2008 until now, Chayes, an expert in discrete mathematics, networks, and game theory, has said very little to the technology and business media about what her teams have been working on—and what it all has to do with Microsoft’s (NASDAQ: MSFT) business. And that’s just fine, because I’m about to tell you.
Microsoft Research New England is three years old this month. Although the full-time research staff has doubled in size since inception, it’s still a fairly small operation—12 staff researchers (including deputy managing director Christian Borgs), another dozen postdocs, a half-dozen software development engineers and support staff, and any number of students and interns. Their research breaks down into six main areas—computer science (including algorithms, cryptography, machine learning and vision, and security), computational biology, economics, mathematics, networks, and social media. (Microsoft declined to say how much it has invested in the lab so far.)
It’s not a typical lineup for a Microsoft research lab, of which there are six worldwide, including the mother lab in Redmond, WA. “Our lab has gone into certain areas that haven’t been nearly as emphasized by the other [Microsoft Research labs]. Most of them are fantastically strong in mainstream computer science disciplines,” Chayes says. “One of the reasons for opening the lab here was because we wanted to help advance the state of the art at the boundaries between computer science and other fields. And Cambridge is a place—of course it’s very strong in computer science—that’s phenomenally strong in other fields.”
Take empirical economics. Susan Athey, the renowned Harvard economist (and Microsoft consultant), heads up a year-old group in the lab that’s focused on analyzing huge datasets that Microsoft has access to—things like online advertising trends and performance, search-engine user behaviors, Xbox Live networks, e-mail and instant-messaging patterns, healthcare trends, and so on. Chayes pitched the idea for the economics group to CEO Steve Ballmer in December 2009, and he said to go for it.
The group’s goal is to find patterns in databases that could make various systems more efficient. Little did they know this would lead to the lab’s first technology to become a Microsoft product. “What an economist does is they try to increase efficiency and social welfare,” Chayes says. “They believe those are the stable points of economic systems.” If you look at the U.S. healthcare system, for example, you will find tons of inefficiency. “The country is being bankrupted by healthcare from a demographic point of view—we’re an aging population, we have Medicare sitting there like a time bomb,” she says.
So a pair of postdocs in the lab, Mark Braverman (now at Princeton) and Mohsen Bayati (now at Stanford), applied some research on complexity theory and machine learning to healthcare data available through Amalga, Microsoft’s health software that’s used by hospitals to synchronize their 80-odd proprietary IT programs. One thing the researchers found was that they could predict which patients would return to the emergency room within 30 days after discharge—a common (and very costly) problem known as “bounceback” that usually means the patient got the wrong diagnosis or prescription.
The Microsoft team looked at all the words the doctors wrote down during the initial hospital visits. By mining the text for key words and phrases—and correlating them with bouncebacks—their software could identify likely bouncebacks and suggest tests to run (or other interventions) that would head off their returning. For example, Chayes says, “If the word ‘fluid’ appears, that’s a huge signal. That substantially increases the probability that someone is going to be readmitted.” (The presence of fluid might indicate an infection, say, which could be prone to misdiagnosis.) The resulting Microsoft product, called “Readmissions Manager,” just shipped as part of Amalga in the last month, she says.
Chayes’s own research in computational biology—such as algorithms for discovering gene regulatory networks, interpreting drug data, and predicting drug targets—could eventually make it into Amalga Life Sciences, a separate program focused on helping scientists make sense of reams of genomic data.
Meanwhile, other projects in the lab’s economics group involve teaming up with Microsoft’s Bing search engine and online services division to analyze things like how foreign currency fluctuations affect ad spending, how social media impacts search behavior and news consumption, how to identify top social influencers (“very important questions if a company like Facebook is going to have a reasonable business model,” Chayes says), and ways to optimize how media companies charge for content across Web, mobile, and tablet platforms. One recent presentation from Athey’s group was titled “Will the Internet Destroy the News Media?”
“It’s very interesting to watch, and it’s actually very disturbing,” Chayes says. “Somebody’s going to have to pay for this content. There’s tremendous use of this content, but the way the money’s flowing is undermining the quality of the content.” (You’re preaching to the choir, Dr. Chayes.)
In another part of the lab, senior researcher danah boyd (she doesn’t capitalize her name) is studying the societal and cultural impact of social media. One example is understanding in a deeper way how young people view privacy. In an era when kids seem to share everything with everybody online, it turns out “privacy to them is their parents not knowing about it,” Chayes says. Another research question is whether the laws and regulations around things like “sexting” and bullying are proportional to the damage they cause, she says—and how to bring the two in line.
Lastly, boyd has been working with Microsoft’s digital crimes unit on ways to identify child pornography online using advanced image processing software. Chayes says boyd has been talking with Twitter and Facebook representatives, as well as Microsoft’s legal experts, about ways that children should be protected (or not) online. And in an interesting sign of an evolving partnership with Facebook, in which Microsoft invested $240 million back in 2007, Microsoft Research has provided its child-pornography-finding software to the social-networking giant, Chayes says.
At heart, Chayes is a researcher and academic leader whose chief concern seems to be making an impact in various fields, and creating an environment where unique connections will continue to be made across disciplines and knowledge advanced. But while it’s still early days for the lab, it also sounds like Microsoft Research New England is doing a fair amount of work that’s directly relevant to the company’s business—and that of its rivals and partners.
It also speaks to tectonic changes going on in the world today around social media, privacy, security, health, and more. What Chayes’s charges learn, and how they bring that knowledge to bear on Microsoft’s products and the world in general, will be fascinating to watch.