Derivatives, gradients, and backpropagation — the mathematics of how neural networks learn from their mistakes.