The Internet community is obsessed with cats, so if you're going to build a neural network consisting of 16,000 computer processors designed to simulate the human brain, then what better task is there than to have it scour the Web for felines? Researchers from Google's X laboratory saw the logic in doing exactly that, and remarkably, the massive neural network actually taught itself to recognize the Internet's favorite type of furball with surprising accuracy.
A detailed report in The New York Times outlines Google's efforts, noting that the newest simulation bests all previous efforts by being twice as capable of picking out objects from a "challenging list of 20,000 distinct items."
In order to successfully learn what a typical cat looks like, Google's research team presented the neural network with random thumbnail images taken from 10 million YouTube videos. The Google brain, as they refer to the network, constructed a hierarchy of memory locations based on the millions of images it saw.
"It is worth noting that our network is still tiny compared to the human visual cortex, which is a million times larger in terms of the number of neurons an synapses," the researchers wrote.