New Theory Cracks Open the Black Box of Deep Neural Networks (wired.com)

I prefer to read this article at the source :)

http://quantamagazine.org