New Theory Cracks Open the Black Box of Deep Learning by Natalie Wolchover via @QuantaMagazine

Naftali Tishby is a researcher with an important new idea about how deep learning might work. He proposes that the most important part of learning is actually forgetting and his ideas about "information bottlenecks" are getting a lot of attention around the web this week.

Wow.  This makes so much sense when you think about it.