Last year, San Francisco startup Jetpac supplied a $ 5,000 purse to anyone who could teach a personal computer to discern which of tens of thousands of vacation photographs were any excellent. Far more than 200 contestants entered. The second-spot winner’s code now lives at the core of JetPac’s iPad travel app, which sifts and displays your Facebook friends’ best travel snapshots to inspire your subsequent excursion.
On Thursday, the organization announced that the final results of the $ 5,000 contest had essentially netted them $ 2.4 million in venture capital funding, primarily from Khosla Ventures and Yahoo co-founder Jerry Yang. That’s some math Jetpac doesn’t require to outsource to appreciate.
Jetpac’s fundraising represents a striking achievement for information science as sport—an method that very first drew focus when Netflix offered $ 1 million to whomever could greatest enhance their movie recommendation algorithm. Ironically, Netflix never ever employed the winning code. Not so at Jetpac.
“Before we had the algorithm, individuals would play with the prototype, and the knowledge was fairly painful,” says Jetpac co-founder Pete Warden, who formerly worked at Apple as an image-processing engineer.
Jetpac ran its data contest by means of Kaggle, a platform for number-crunching competitions that has built some critical geek cachet in the course of its short existence. NASA sponsored a competition to map the universe’s dark matter. Facebook backed a nerd throwdown in which the prize was a job interview at Facebook. PayPal mafioso Max Levchin is chairman of the board.
Kaggle hosts public competitions, such as Jetpac’s, in which anybody can attempt to resolve the difficulty posed by the contest sponsor. Top competitors get invited to Kaggle’s private contests, which typically involve proprietary business information and bigger purses.
The beauty of solving information difficulties via contests is that like other sports, the final score speaks for itself. Math tells you which algorithm offers the most correct outcomes. It’s a pure meritocracy.
Take Jetpac, which prior to the contest used humans to rate the top quality of 30,000 shared pictures. Contestants “trained” their algorithms utilizing meta-information associated with 10,000 of the pictures, then ran their code on the remaining 20,000 to see how well they’d accomplished. The purpose: Develop an algorithm that could rank the quality of the photographs in a way that most closely matched the subjective human ratings just by analyzing items like captions, photo size and location.
The algorithm Jetpac ultimately employed came from the second-spot finisher, whose code Warden says was nearly as correct as the winner’s but was considerably cleaner and less difficult to drop proper into the company’s personal software. Caption words that indicated a bad vacation photo included “mommy,” “graduation,” “CEO” and “San Jose.” The top word indicating a good photo was “tomb.” (Warden says believe European cathedrals and Angkor Wat.) The prize for second location: $ 1,500.
That amount pales subsequent to $ two.4 million, but Warden says the competitor, an Oxford physics Ph.D. who operates days for a London-based investment fund doing “algorithmic trading,” doesn’t lose out completely. Jetpac only gets a non-exclusive license to the algorithm, per Kaggle guidelines—though that shouldn’t difficulty the organization’s new backers, because Khosla also funds Kaggle.
The winner, Jason Tigg, who’s ranked fifth overall among all Kaggle competitors, known as Jetpac’s achievement “great news.”
“They did all the hard operate and then allowed me and other Kagglers the opportunity to do the enjoyable bit,” Tigg says.
Warden says the Kaggle knowledge was so effective that Jetpac plans to sponsor a lot more contests with its new money, which includes an effort to analyze the photos’ visual data to see if an algorithm can judge a image by “looking” at it. The contests give startups access to a level of smarts they otherwise couldn’t afford, he says, considering that few correct data geeks want to deal with the non-math–related headaches that come with startup life.
“They enjoy the actual puzzle solving aspect of the machine understanding,” Warden says. “But they don’t like all the other stuff that comes with a job.”
