Smart sensorsTeaching sensors to think for themselves
There is a major problem with sensors: data overload; as sensors gather more and more information, it has become increasingly difficult for human users to separate out what is relevant from what is not; two U Vermont researchers received a grant from DARPA to teach sensors what to look for — and what not to look for
The Defense Advanced Research Projects Agency (DARPA) Mathematics of Sensing, Exploitation, and Execution (MSEE) program has awarded $500,000 to Drs. Joshua Bongard and Christopher Danforth, assistant professors in the University of Vermont College of Engineering and Mathematical Sciences, for their research to find a way to teach sensors how to think. The mission of DARPA is to pursue and exploit fundamental science and innovation for national defense.
DARPA’s newest project aims to address a major problem of sensors: data overload. As sensors gather more and more information, it has become increasingly difficult for human users to separate out what is relevant from what is not.
“[There are] all those sensors out there that are providing us with huge streams of information but 99 percent of the information that’s coming back is not useful,” Bongard told the Vermont Cynic.
Over the next three years, Bongard and Danforth will work to fix this problem by creating a filtering mechanism that will integrate with sensors in order to make them more refined and ultimately, better machines. “As we start to deploy more sensors out into the world, we don’t want more and more information; we want the same amount of information, but better focused,” Bongard said.
Bongard notes that there are two problems involved in teacing sensors how to think: teaching the what to look for – and teaching what not to look for. “The problem is that if you look at a big enough set of numbers, there are an infinite number of interesting patterns in there that just happen by chance,” Bongard said. “It’s not so much getting something to recognize patterns, but ignoring all the irrelevant patterns. That will be the biggest challenge,” he told the Cynic.
