Expand the neural network code

[qzhu2017]

Currently, we have a single neuron code, but it would be good to expand it 2 neurons, 2 layers. This way, we will have a better understanding on the back-propagation mechanism used in the updates of the weights.