The Stilwell Brain

This comes from [the video](https://www.youtube.com/watch?v=rA5qnZUXcqo&pbjreload=10) by Vsauce. I know you've already made mnist classification example, but this is quite different. The idea is creating a neural network model with 5 layers almost like it is in a human brain: retina, v1, v2, v4 and IT-layer.

* Retina receives inputs (pixels of handwritten digit)
* V1 neurones are sensitive to lines, which means that a neurone fires if three of retina layers make a line
* V2 neurones are sensitive to angles
* V4 are sensitive to shapes
* IT is pretty much like an output softmax layer

I think this algorithm is worth implementing because it gives deep understanding of how human brain works and how it is related to artificial neural networks.