Do You Like Funny Cat Videos

When com­puter sci­en­tists at Google’s mys­te­ri­ous X lab built a neural net­work of 16,000 com­puter proces­sors with one bil­lion con­nec­tions and let it browse YouTube, it did what many web users might do — it began to look for cats.

The “brain” sim­u­la­tion was exposed to 10 mil­lion ran­domly selected YouTube video thumb­nails over the course of three days and, after being pre­sented with a list of 20,000 dif­fer­ent items, it began to rec­og­nize pic­tures of cats using a “deep learn­ing” algo­rithm. This was despite being fed no infor­ma­tion on dis­tin­guish­ing fea­tures that might help iden­tify one.


Pick­ing up on the most com­monly occur­ring images fea­tured on YouTube, the sys­tem achieved 81.7 per­cent accu­racy in detect­ing human faces, 76.7 per­cent accu­racy when iden­ti­fy­ing human body parts and 74.8 per­cent accu­racy when iden­ti­fy­ing cats.


Leave a reply