Backpropagation calculates gradients by applying the chain rule...
L1 regularization adds absolute value of weights to loss function...