Backpropagation
Tracing the error at the output backward through the network, adjusting each weight in proportion to how much it contributed to that error. The lecture explains it conceptually (the relay-team analogy) and gives no formulas.
When to use
It is step 4 ("Backward") of the training loop: initialise, Forward Pass, Loss Function, backward, then update the weights with Gradient Descent. In an exam, name it as the credit-assignment step rather than deriving it.
See Lec 8.