The computer is based on a neural network of 16,000 processing cores with more than a billion interconnections, each very roughly simulating a connection in a human brain.

A team from Googles cutting-edge research lab, Google X, and Stanford University, fed the system 10 million thumbnail images taken from YouTube as training and then tested whether it was able to recognise 20,000 objects in new images.

It performed more than twice as accurately as any previous neural network, The New York Times reports. Among the objects the system learned to recognise was a cat, one of the most regulars star of viral clips uploaded by YouTube members.

We never told it during the training, This is a cat, said Google fellow Dr Jeff Dean. It basically invented the concept of a cat.

Overall, the neural network achieved 15.8 per cent accuracy. As well as cats faces, it learned the concepts of human faces and bodies, by compiling a ghostly image of their general features.

The research differed from how most artificially intelligent systems are trained, in that it was given no help by human supervisors labelling features.

The idea is that instead of having teams of researchers trying to find out how to find edges, you instead throw a ton of data at the algorithm and you let the data speak and have the software automatically learn from the data, said Andrew Ng of Stanford University.

Neural networks have myriad potential applications, including in speech recognition software and image search, which Google is working to improve.

The firms secretive Google X Lab works on cutting-edge technologies such as self-driving cars and augmented reality.

See the rest here: