Microsoft Research’s Jennifer Chayes: 5 Projects for the Future of Computing

4/9/12Follow @gthuang

(Page 2 of 3)

human beings do, and it is going to understand what we want and how to present it. And each person’s experience with their devices is going to be unique to that person. We don’t have that now. But I think five to 10 years from now, with machine learning, we will have that,” Chayes says. Computers “will understand where to pull in these various threads” so the right information is delivered to us and “we really don’t have to think about it.”

As for Microsoft’s not getting as much media love as other tech darlings, Chayes downplays the competition and focuses in-house. “We have been getting a lot of attention around Xbox and Kinect,” she says. “That’s our new cool thing, our new shiny toy, which is a lot more than a toy. I think it will be in every aspect of our enterprise [business] as well as our consumer and the living room and all that. But, beyond that, machine learning is becoming such a big part of what we do.”

And with that, let’s take a look at five projects from Microsoft Research New England that exemplify what Chayes is talking about—and could lead to some interesting new products (and possibly help shape the future of computing):

1. Machine learning for the cloud. This is a project led by postdoc Ohad Shamir together with researchers in Redmond. The basic idea is to use machine learning algorithms to help startups and other organizations bid properly for cloud-computing resources. “They can either do spot pricing or they can be buying upfront at higher prices—how do they optimize that?” Chayes says. “For us, on the cloud provider side [with Microsoft Azure], it would give us an analysis of what the flow was of the kinds of requests coming in, so that we could time things better and use our energy resources better. There will be opportunities for data markets in the cloud that will be absolutely huge.”

This may not sound very sexy, but neither did Amazon Web Services back in 2006. “If you make it cheaper and easier for a startup to get the cloud resources it needs, that’s not your shiny object, but to a startup business, it’s everything,” Chayes says.

2. Machine learning for people and tasks. This one is more along the lines of what people have been talking about as “machine learning” for the past 20 years. Actually it’s two projects. The first has to do with categorizing which images are similar to one another—something people are great at, but machines stink at. Lab members Adam Kalai, Ce Liu, Ohad Shamir, and their collaborators used crowdsourcing through Amazon’s Mechanical Turk to teach a machine how to decide whether image A—a floor tile, national flag, or human face, say—is more similar to image B or image C. The science has to do with understanding how humans perceive similarities, and incorporating those judgments into a machine. The applications could include e-retailers displaying things like home furnishings or apparel in a way that lets you drill down to styles you like by clicking on images, rather than sorting items just by their color or other blunt tags.

The second project has to do with “programming by example.” Led by Kalai, Microsoft technical fellow Butler Lampson, and senior researcher Sumit Gulwani, this one involves … Next Page »

Gregory T. Huang is Xconomy's Deputy Editor, National IT Editor, and the Editor of Xconomy Boston. You can e-mail him at gthuang@xconomy.com or call him at 617-252-7323. Follow @gthuang

Single Page Currently on Page: 1 2 3 previous page

By posting a comment, you agree to our terms and conditions.

  • http://pharmaceuticalintelligence.wordpress.com Aviva Lev-Ari, PhD

    Start-ups lacking a value proposition squarely in one of these areas face an arduous and unsatisfying trek along Sand Hill Road. Successful entrepreneurs explicitly need a Zen understanding of how data drives business value in their target market. Successful products explicitly need to be able to help extract the value buried in ever-larger quantities of data. And successful pricing models explicitly need to be founded on delivering value from the semantics of data. Period.

    http://www.forbes.com/sites/ciocentral/2012/03/12/big-data-the-only-business-model-that-tech-has-left/2/

  • Pingback: Machine learning to get its day in the Redmond sun | ZDNet

  • Pingback: Machine learning to get its day in the Redmond sun | Technology News

  • Pingback: Microsoft Adds NYC Research Site Targeting Social Science Questions | Xconomy

  • Guest

    I have no doubt in my mind that Cambridge is a great place for Microsoft to do R&D. The next great centers of innovation will be in major cities and hubs of innovation (especially those near universities). The model of doing R&D in remote office parks where land is cheap just hasn’t been as productive in terms of innovation as we would like.

    With that said, I find Microsofts 5 ideas here to be somewhat important but not the kind of thing you’d put on the cover of a magazine. Compare what Microsoft is doing to Google (self driving cars), 23andMe (genomic research and services), IBM (distributed computing for studying proteins, vaccines, etc… with the World Community Grid) and of course Apple’s Siri.

    Microsoft has some great projects like Robotics Studio that could be much further developed in terms their capabilities and applications with real wow factor and market potential. When it comes to new technologies it would make sense for Microsoft to go further into hardware. The reason being is that in the software space, Microsoft has a great deal of competition from startups. When it comes to hardware, R&D is far more capital intensive, an area where Microsoft has a major advantage over the little guys.