Google Uses 1,000 Computers To Simulate Human Brain
This article was written by Amit Chowdhry. You can follow me at @amitchowdhry or on Google+ at +AmitChowdhry
The research team at Google stitched 1,000 computers together with 16,000 cores to create a neural network that has over 1 billion connections. They used this super-computer to search for cats on YouTube. When the super-computer was presented with 10 million images from random videos, it was able to teach itself what cats look like without assistance. The key take-away here is that computers have the ability to learn face detection without labeled images. There was about a 15.8% accuracy when recognizing 20,000 object categories.