It only took 16,000 computer processors to do it, but over the span of just a few years a group of Google scientists were able to simulate how the human brain identifies things they find on the Internet. More specifically, the neural network was able to train itself to recognize cats. The way it was able to recognize them actually reflected biological theories where objects are identified by trained individual neurons in the brain.
After being exposed to a few million digital images of cats from videos on YouTube, the neural network tapped into its memory of what it extracted and learned from the images before putting together its own image of a cat. Like humans, it was able to understand the general features of the cat through repeated exposure to the images.
While an actual human’s ability to identify cats on the Internet is not exactly impressive by any means, Google’s simulation experiment suggests that machine advancements are getting that much closer to human-like functions, and are leading to machines being able to better visibly see and perceive things, understand human speech and translate languages.
[via New York Times]