Google's Neural Network Learns to Find Cats on YouTube
Google hooked up 16,000 computers so that they could learn to identify cats in YouTube videos (like this one).
Computers have finally discovered the joy of looking at cats in YouTube videos — a 21st-century pastime shared by millions of people around the world. What's really impressive is that the computers had no help from humans in learning to identify our aloof feline friends.
Google's X Lab scientists kicked off the artificial intelligence journey of self-discovery by connecting 16,000 computer processors to make a huge neural network "brain." They then turned the neural network loose on thumbnail images taken from 10 million YouTube videos for three days, according to the New York Times.
The neural network had to learn how to identify distinct objects such as cat faces without having ever been trained on labeled images. It proved wildly successful by computer standards with an almost 16 percent accuracy in recognizing 20,000 object categories — about a 70 percent leap in improvement over previous machine learning performances.
Such success could eventually boost Google's search engines by helping them automatically identify what people are searching for or looking at on the Internet. Anyone curious about the details of Google's latest work can read the advance paper published ahead of a presentation at the International Conference on Machine Learning being held in Edinburgh, Scotland from June 26 to July 1.
Source: New York Times