If we know which kind of output we want to have and if each neuron's output is a smooth function of its input, then the change in weights to get the right output from the input can be computed using calculus.

Following this strategy, we get backpropagation⇒