Winning the DARPA Challenge: Crowdsourcing, algorithms and detective work
It’s easy to cut a piece of paper into little bits. The trick is putting it back together again to read what’s on it. Now imagine doing that on a battlefield.
That was the goal of the Defense Advanced Research Agency’s Shredder Challenge, which was won this month by a small San Francisco-based team that correctly reconstructed the event’s five documents and solved their related puzzles.
The “All Your Shreds Are Belong to U.S.” team developed custom-coded, computer vision algorithms that recommended fragment pairings to human assemblers for verification.
System would monitor feds for signs they're 'breaking bad'
DARPA's new cyber tack: act like a hacker
The team spent nearly 600 man-hours creating the algorithms and piecing together the documents that were shredded into more than 10,000 pieces. Some 9,000 teams registered for the Challenge, DARPA officials said.
The most effective approaches in the event combined computational and crowdsourcing, with a dash of detective work thrown in, Dan Kaufman, director of DARPA’s Information Innovation Office, said in a statement. It took the winning team 33 days to solve all five puzzles and win the $50,000 prize.
DARPA’s Shredder Challenge is a preliminary look at potential technologies and applications that could help warfighters in the field. The agency is looking for capabilities that would allow soldiers to quickly reconstruct and pull information from shredded or destroyed documents. DARPA also wants to understand the potential vulnerabilities the U.S. government faces when it shreds sensitive documents and hopes to apply those lessons to new information security techniques.
Agency personnel will meet with the winning team to learn more about the techniques they used to solve all five puzzles. However, DARPA has no plans at this time to develop software related to the Shredder Challenge, said agency spokesman Eric Mazzacone. He added that any intellectual property rights remain with the participants.