Posted by Rob Knies

TechFest 2013 logo

You might have heard of Drew Purves before. As head of the Computational Ecology and Environmental Science Group at Microsoft Research Cambridge, he has gained attention for a number of his projects, from modeling forest dynamics to FetchClimate, that use big data and build models to make predictions about future environmental conditions.

He’s back at TechFest this year with a different kind of project. Called Predictive Decision-Making at the Speed of Thought, this effort provides the fundamental research needed to build predictive models for many different types of data, not just the environmental. The goal is to generalize the approach he has been pursuing to make such tools available to a broad range of organizations and businesses.

“In the environmental sciences where we work, but also much more broadly, there’s a lot of demand for the pipeline that goes from big data through models to predictions of important things,” Purves explains. “We know, fundamentally, how to do that, but the technical barriers at the moment are so high that it’s the domain of specialist experts, which, in turn, means that it’s only the world’s largest organizations that can afford to support that kind of data-to-prediction pipeline.”

Purves and his colleagues have developed a browser application that shares their learning in a manner that can be used by the many, instead of just the few.

“It also allows us to push the model into Windows Azure,” he says, “so the model then can be run in Azure on demand whenever and wherever it’s needed, without ever leaving the browser. You can literally go through that process about as quickly as you can describe it. We’re looking at something that would’ve taken weeks, and the demo takes about three to five minutes to begin with data and end up with a model.”

Given the interest in sharing such predictive work more broadly, TechFest plays a key role in popularizing the concept.

Get Microsoft Silverlight

“Ever since I’ve been in the group,” Purves says, “it’s had this two-part mission: to do the fundamental research to generate the predictive models we need in environmental sciences on one hand, and, on the other hand, to look for opportunities to develop software that accelerates that work for us—and therefore accelerates that kind of work for others.

“The things involved in this tool—getting data, supplementing that data with environmental information, defining a model, training the model against the data, and making predictions—that’s what we’re all about in the environmental sciences, but it’s a much more general story. These tools are ones we want to share with our academic colleagues doing the kind of work they do, but here at TechFest, it’s an opportunity to offer them up to Microsoft to package them much more broadly, because there’s a much more general story here.”

Moving from domain-specific scientific exploration to model-based generalization does necessitate a certain change in direction, but Purves embraces the challenge.

“Speaking personally, I like to use all different parts of my brain,” he says. “There’s a certain kind of hard-thinking, raw, fundamental science that we’re pushing. This is on the other end of the spectrum, where we’re thinking about software and what should be the name and what should it look like, what should be the user experience—even here at TechFest, during the traveling-salesman part of pitching it. I like that full spectrum of all of that. I like making the web page and the YouTube video.

“On a slightly more serious level, one way to have impact in terms of the environmental sciences is to carry out some science, write that up in a traditional paper, and see that taken up and inspire other scientists. But quite another way to have an impact is to give them tools to accelerate that science.

“If you get that right, you might be able to have a much larger impact. If we can get the Microsoft machine involved, to raise that software to the standard of the kind of software that Microsoft produces, then we could really have a dramatic impact on huge, whole swaths of environmental science in a way that a traditional scientific career just basically can’t.”