This was written by the 3 biggest names in the field back in 2015, right after the breakthrough success of neural networks for computer vision tasks, just before the AlphaGo paper came out. It's math-y in places, but interesting to read even you don't want to try to follow the math.