Google Brain watches Youtube, recognises cats.
25th June 2012
One of the many experimental projects at the Google X laboratory was a "deep learning" technique whose aim was to improve the adaptability of machine learning technology, especially when dealing with unlabelled data (data that has not been labelled and identified by humans). The project was later named Google Brain.
Using 16,000 CPUs in their data centres and software that let it easily break programs into many tasks that could be computed simultaneously, the project was able to teach itself to recognise patterns in random samples of unlabelled data.
After processing randomly-selected youtube thumbnails, Google Brain produced a number of images that it had learned to recognise. One of these images was the face of a cat.
The scientists behind the project said that it appeared they had developed a simulation of what takes place in the brain’s visual cortex.