No shame in Baidu's image recognition software game, the headlines shouted when the Chinese search company beat out Google in a standardized accuracy test. But in truth, there was so much shame.
Baidu was forced to issue an apology when it was revealed the company had engaged in a vast scheme to cheat the system. The test's object was to get the best software score against the ImageNet Challenge. The rules state your code can be tested only twice a week to mitigate the element of chance. Baidu tested theirs four times as much as the rules allowed using multiple email accounts, stacking the odds in their favor.
Why cheat on a little-known academic test? Because that’s the kind of inside baseball scoring that matters to the geeks software giants have been recruiting for their research groups. As Technology Review puts it, “A handful of standardized tests developed in academia are the currency by which these research groups compare one another’s progress and promote their achievements to the public.” And its a good way to show off to the engineers you want to recruit for a six-figure job.
Now if you’ll excuse me, I’ll be renewing my login at Code Academy.
