Researchers Devise Algorithms To Prevent Information Overload
Pulling information from vast sensor arrays is easy. Sorting it and deciding what's useful without being overwhelmed is the hard part.
Hundreds of video cameras lace the London subway system, capturing the faces of passing commuters. Suppose one of them catches the visage of a bombing suspect. Then what?
A group of researchers at the University of Illinois at Urbana-Champaign believes there's a way for software to govern a network of sensors so that such valuable information rises above the mass of random data.
- Government Analytics: Set Goals, Drive Accountability and Improve Outcomes
- 2012 IBM Chief Information Security Officer Assessment
Public video cameras, RFID-enabled warehouses, and mixture monitors in a chemical plant are all examples of extensive sensor networks that may one day feed enormous amounts of information into self-monitoring systems that will pluck useful kernels from the chaff.
"Getting the information we need is not the problem; sorting it and deciding what is useful without being overwhelmed is the challenge," says Robert Ghrist, associate professor of mathematics at the University of Illinois. But Ghrist and a team of researchers at Illinois and seven other universities, along with the Bell Labs research unit of Lucent Technologies, propose to do just that.
Ghrist, a researcher of topology mathematics at the university's Coordinated Science Laboratory, will co-lead research for Stomp, or the Sensor Topology & Minimal Planning project, which won $8 million in funding from the Defense Advanced Research Projects Agency on Oct. 5. Topology is the study of abstract spaces.
When asked about the science's possible business applications, Ghrist cites a hypothetical example of detecting holes in the coverage of a cell phone network. Topology mathematics can map the twists and curves of the holes.
Once topology has captured sensor information showing where holes exist, it can map them out in a way that provides the guidance needed to fix them, he says. In short, it's about integrating local readings from many sensors "into a global picture."
Yuliy Baryshnikov, lead engineer on the research for Bell Labs, used the example of a grid of 1,600 motion sensors, each occasionally offering irrelevant feedback such as leaves dropping from trees. In a display of such continuous feedback, a human eye may miss intruders proceeding through a monitored area. But topology algorithms resulting from the research will spot them, he says.
The "minimal planning" part of the project reflects its goal of building the smallest sensor network necessary to get a job done, as opposed to overinvesting in sensor placement.